Context-Aware Assistive Systems for Augmented Work
Transcription
Context-Aware Assistive Systems for Augmented Work
Context-Aware Assistive Systems for Augmented Work. A Framework Using Gamification and Projection Oliver KOrn CONTEXT-AWARE ASSISTIVE SYSTEMS FOR AUGMENTED WORK. A FRAMEWORK USING GAMIFICATION AND PROJECTION Von der Fakultät für Informatik, Elektrotechnik und Informationstechnik der Universität Stuttgart zur Erlangung der Würde eines Doktors der Philosophie (Dr. phil.) genehmigte Abhandlung vorgelegt von OLIVER KORN Erstgutachter: Prof. Dr. Albrecht Schmidt Zweitgutachterin: Prof. Dr. Fillia Makedon, PhD Tag der mündlichen Prüfung: 21.05.2014 Institut für Visualisierung und Interaktive Systeme (VIS) der Universität Stuttgart 2014 Abstract i ABSTRACT While context-aware assistive systems (CAAS) have become ubiquitous in cars or smartphones, most workers in production environments still rely on their skills and expertise to make the right choices and movements. The quality gates currently estab lished by the industry successfully remove “waste” i.e. failed work results from the workflow, but they usually operate in a spatial and temporal distance from the work place and the worker. Thus workers lack the opportunity to learn from problems on the fly and to improve their work habits. Today’s motion recognition systems already allow to continuously analyze work ac tions. The corresponding middleware can interpret these actions in real-time and micro projectors allow to display work-relevant information directly into a worker’s field of vision. Thus the technical pre-requisites for CAAS with instant feedback directly at the workplace essentially have been established. Although every worker would benefit from context-aware assistive systems, impaired persons and elderly persons with reduced physical and mental capabilities require such systems the most. CAAS have the potential to empower these persons to do more complex work or to remain in regular jobs longer. Thus they combine economic benefits with inclusion and address the demographic change. After an overview of the relevant backgrounds from ethics, psychology, computer science and engineering as well as the relevant state-of-the-art, we establish require ments which result in a model for ideal CAAS. As the framework aims to improve not only the work results but also the work process and thus the workers’ motivation, the model incorporates elements from game design. This process of “gamifying” realworld actions is called gamification. We describe an exemplary implementation covering essential aspects of the model. The effects of both the augmentation by projection and by gamification are evaluated with impaired persons in a sheltered work organization using empirical methods com mon in human computer interaction. An additional focus lies on the ethical implica tions of assistive systems which supervise and model their users in real-time. While this thesis represents a starting point for research on CAAS in workplaces, im portant aspects like real-time error tracking or the integration of emotion detection are still subjects of future research. We hope that future CASS based on this work will help to make work not only more productive but also more enjoyable and motivating for workers. ii Zusammenfassung ZUSAMMENFASSUNG Während kontextbewusste Systeme (context-aware assistive systems: CAAS) in Smartphones und Autos allgegenwärtig geworden sind, verlassen sich Arbeiter in Pro duktionsumgebungen meist noch auf Fähigkeiten und Erfahrung, um Entscheidungen zu treffen und Arbeitshandlungen richtig durchzuführen. Die derzeit in der Industrie etablierten Qualitätsprüfungen entfernen zwar erfolgreich Ausschuss (d. h. fehlerhafte Produkte) aus dem Durchlauf, werden jedoch meist in räumlicher und zeitlicher Dis tanz vom Arbeitsplatz betrieben. Daher fehlt den Arbeitern die Möglichkeit, während des Arbeitsprozesses aus Problemen zu lernen und ihr Arbeitsverhalten zu verbessern. Aktuelle Bewegungserkennungssysteme erlauben es bereits, Arbeitshandlungen kon tinuierlich zu analysieren. Die dazugehörige Middleware kann diese in Echtzeit aus werten und Projektoren erlauben es, arbeitsrelevante Informationen direkt in das Blickfeld des Arbeiters projizieren. Damit wurden die technischen Voraussetzungen für arbeitsplatzintegrierte CAAS mit Feedback in Echtzeit geschaffen. Obwohl jeder Arbeiter von solchen kontextbewussten Systemen profitieren würde, ist der Bedarf bei Menschen mit Behinderungen sowie älteren Berufstätigen mit redu zierter körperlicher und geistiger Leistungsfähigkeit am größten. CAAS haben das Potenzial diesen Anwender zu ermöglichen, komplexere Arbeiten zu verrichten oder ihrer regulären Arbeit länger nachzugehen. Damit kombinieren CAAS wirtschaftliche Vorteile mit Inklusion und adressieren den demographischen Wandel. Auf Basis einer Übersicht von Grundlagen aus den Bereichen Ethik, Psychologie, In formatik und Maschinenbau sowie des relevanten Standes der Technik werden An forderungen abgeleitet, die in einem Modell idealer kontextbewusster Systeme resultieren. Da diese Systeme darauf abzielen, nicht nur die Arbeitsresultate sondern auch die Motivation der Arbeiter zu steigern, integriert das Modell Elemente aus der Computerspieleentwicklung. Dieser Prozess wird Gamification genannt. Die exemplarische Implementierung der essentiellen Aspekte des Modells wird be schrieben. Die Auswirkungen der Augmentierung durch Projektion und durch Gami fication werden mit behinderten Menschen in einer beschützenden Werkstätte evaluiert. Hierbei werden übliche empirische Methoden der Mensch-Computer-Inter aktion angewandt. Ein zusätzlicher Fokus sind die ethischen Implikationen von As sistenzsystemen, welche ihre Anwender in Echtzeit überwachen und modellieren. Diese Arbeit stellt einen Anfangspunkt für die Forschung zu CAAS im Arbeitsumfeld dar. Wichtige Aspekte wie Fehlererkennung oder Emotionserkennung in Echtzeit sind noch immer offene Forschungsfragen. Wir hoffen, dass zukünftige CASS, die auf die ser Arbeit aufbauen, dabei helfen, Arbeit nicht nur produktiver sondern auch ange nehmer und motivierender für die Arbeiter zu gestalten. Acknowledgements iii ACKNOWLEDGEMENTS In the past three years I had the opportunity to work with a number of exceptional people. The following list is not exhaustive but mentions people I worked closely with. At first I want to thank my supervisor Albrecht Schmidt. He has been a constant supporter and enabler. Albrecht’s idea of leading by giving freely has been an inspi ration to me that goes well beyond academic honors. My second supervisor Fillia Makedon from the University of Texas Arlington is one of the founders of the field of assistive technologies. Her support for my work began when we first met at a PETRA conference in Crete and continues ever since. If I still cannot perform a proper Sirtaki it is entirely due to my own shortcomings. The foundations for the use of motion recognition in assistive contexts were estab lished during the BMBF-funded project motivotion60+ at the Korion GmbH (2009 2012, framework Ambient Assisted Living). I want to thank my friend and colleague Robert Konrad for hacking the first Kinect and being the most reliable programming genius I know. His basic findings on motion recognition are still used in this work. This also applies to Bodo Runde whom I want to thank for helping to develop the first motion-based exergames for elderly people. The gamification component used in the assistive system would not look and work the way it does without the help of Valentin Schwind, an excellent graphic designer and developer. Much of the research presented in this work is related to the funded project ASLM (assistive systems for persons with impairments working in assembly). It was acquired by Thomas Hörz from the University of Applied Sciences Esslingen (HE), where I spent my first year with basic research on motion recognition and assistive technolo gies in production. I want to thank him for giving a computer scientist access to the Faculty of Engineering; it is also thanks to his industriousness that a patent application on the algorithm used in the system was filed and that we received the Gips-Schüle Award for applied research in the field “Human and Technology” in 2013. Many chapters of this thesis are based on the hard work of excellent students. I espe cially want to thank Björn Böhmert, Daniel Kaupp and Benjamin Knapp. Björn worked with me in the basic research on motion recognition and did a great job in implementing the “adjoined spheres model” for observing 3D areas in workspaces. Daniel worked on the development of the “Assistive Systems Experiment Designer” and assisted during the field studies. Benjamin also helped during the studies and de veloped a tool that automatically analyzes the experiments’ extensive log files. Finally Adrian Rees did an excellent job in proof-reading. The extensive studies would not have been possible without the support of the shel tered work organization BWH in Heilbronn. I want to thank its CEO Alfred Grimm iv Acknowledgements for giving management support and supervising the study’s ethical compliance as well as Christof Sanwald and Jürgen Kleiner for giving valuable expert input in the de sign phase and heart-warming on-site support during the study. The assembly table used during these studies was constructed and built by the Schnaithmann GmbH. I want to thank its CEO Karl Schnaithmann and the project team: Volker Sieber, Daniela Bühler, Bernhard Bader and Marc Hentzner. Seven student project groups had to endure my supervision at the HE. They provided market data, tested and integrated hardware or sensors and conducted basic studies on competence levels or object complexity. I want to thank them for their dedication and good work: Tobias Scheffel, Stefan Leyh, Stephan Kazmierczak, Fabian Dörre, Christian Löffler, Max Korn, Roman Rinas, Adrian Starczewski, Julian Besserer, Tobias Reim, Johannes Hirneise, Julian-Matthias Gaiser, Simon Schweik ert, Philipp Sigle, Torben Kamphans, Marc Dellenbach, Thilo Anhorn and Axel Baier. In the last phase of the ASLM project it is thanks to my former colleague Manuel Kölz that the subsequent student groups were perfectly supervised. In February 2013 I continued my research at the University of Stuttgart where I had the pleasure of receiving valuable advice for the final stage of my thesis from fellow PhD students. Stephan Abele gave valuable support for the statistical analysis. At Albrecht’s Institute for Visualization and Interactive Systems (VIS) my colleague Markus Funk worked closely with me. I was inspired from his straightforward char acter and benefited from his technical expertise in motion and object recognition. The large-scale federal project motionEAP funded by the Federal Ministry for Eco nomic Affairs made it possible to continue the research seamlessly and also supported the print version of this work. The project was encouraged by my long-standing men tor Thomas Wahl, the vice-president of DLR’s project section. Thanks for supporting me since my first steps in project work in the early noughties at Fraunhofer. I want to thank my parents Elfi and Ulrich Korn for many things – but especially for not insisting on prohibiting computer games. The days and nights spent in front of the ZX81 and the Commodore Amiga laid the foundation of the fascination for computers and software that still fuels my business and research interests today. I also want to thank my mother in law Gisela Hainbuch for selflessly supporting me and my family during the last years: you have been our personal assistive system. Finally I want to thank my family: my wife Silja Korn, my daughter Helena and my son Leander. Completing a PhD in the late thirties is a joint effort. Thanks Silja for letting me spend dozens of nights at the lab while caring for the kids alone. Thanks for enduring numerous conference travels and sorry for all the evenings and weekends spent writing papers or this text. You are the light of my life. Table of Contents v TABLE OF CONTENTS 1 Introduction 1.1 1 Motivation 1 1.1.1 Automation versus Manual Assembly 2 1.1.2 Increased Demand for Impaired Workers 3 1.1.3 Demographic Change 4 1.1.4 Pursuing the Ethical Good 4 1.2 Thesis Outline 5 1.3 Research Questions 7 1.4 Methodology 8 1.4.1 State-of-the-Art and Requirement Studies 8 1.4.2 Prototypes 8 1.4.3 Evaluation 9 1.5 Research Contributions 9 1.6 Publications 9 Background 12 2.1 Related Disciplines 12 2.2 Demography and Targeted Users 14 2 2.3 2.4 2.5 2.6 2.2.1 Elderly Persons 14 2.2.2 Disabled and Impaired Persons 16 2.2.3 Aging and Disability 20 2.2.4 Sheltered Work and Supported Work 21 Ethical Dimensions 23 2.3.1 23 Value of Technology 2.3.2 Value of Work 24 2.3.3 Implications for the Work of Persons with Disabilities 26 Psychology 27 2.4.1 HAAT-model 27 2.4.2 Cognitive Workloads 29 2.4.3 Flow 30 Computer Science 33 2.5.1 Human-Computer Interaction 33 2.5.2 Assistive Technology 35 2.5.3 Augmented Reality 37 2.5.4 Artificial Intelligence 38 Engineering and Production Management 40 2.6.1 Measuring Work Processes with MTM 40 2.6.2 Assembly Workplaces 42 vi Table of Contents 2.6.3 3 Digital and Virtual Factory, Cyber-physical Systems Related Work 44 48 3.1 Motion Recognition 48 3.2 Projection 51 3.3 Gamification 53 3.4 Assistive Systems at Workplaces 57 3.4.1 Assistive Systems Using Motion Recognition 58 3.4.2 Assistive Systems Using Projection 60 3.4.3 Assistive Systems Using Gamification 62 3.5 Ethical Standards for Assistive Technology 65 Requirements 67 4.1 Standards for Designing Interactive Systems 67 4.2 Constraints and Adaptations 69 4.2.1 69 4 4.3 5 Focus of Interaction 4.2.2 Total Quality 70 4.2.3 Universal Design 71 Requirements Study 74 Model and Architecture 78 5.1 Adapted HAAT-model 78 5.2 Framework for an Adaptive Game System 80 5.3 Flow Curves 82 5.4 Model for Context-Aware Assistive Systems 83 Implementation 86 6.1 System Overview 87 6.2 Physical Integration in the Work Environment 87 6.3 Motion Recognition 89 6.3.1 Technical Restrictions and Solution 89 6.3.2 The Adjoined Spheres Approach 90 6 6.4 6.5 7 Instruction and Performance Analysis 93 6.4.1 Visual Output and Projection 94 6.4.2 Designer Component 96 6.4.3 Runner Component 100 Gamification Component 103 6.5.1 Designing Flow 103 6.5.2 Implementing Flow 105 Studies and Evaluation 110 7.1 Participants and Preparatory Study 110 7.2 Procedure 114 Table of Contents 7.3 vii 7.2.1 Questionnaire Design 114 7.2.2 Assembly Task 117 119 Apparatus 7.3.1 Time Measurement 7.3.2 Quality measurement 119 121 7.4 Data 7.5 Experiment Results and Discussion 128 7.5.1 Overall Results 128 7.5.2 Analysis of Time 129 7.5.3 Analysis of Errors 132 7.5.4 Sub-Populations 135 7.6 123 Questionnaire Results and Discussion 137 7.6.1 Pre-Experiment Results 137 7.6.2 Generic Post-Experiment Results 139 7.6.3 Specific Post-Experiment Results 144 7.7 Qualitative Findings 148 7.8 Discussion 150 8 7.8.1 Similarities between the Scenarios 150 7.8.2 Differences between the Scenarios 151 Conclusion 153 8.1 Summary of Research Contributions 153 8.2 Ethical Implications: Towards Humane Work 161 8.3 Future Research 163 8.3.1 Technical Perspectives 163 8.3.2 Ethical Perspectives 165 9 Supplement 166 9.1 Implementation Details 166 9.2 Questionnaires in Detail 168 9.3 Study Result Details 174 10 References 181 11 Index 191 viii List of Figures LIST OF FIGURES Figure 1: Relation between automation, lot size and the application of CAAS........................... 2 Figure 2: Disciplines related to context-aware assistive systems............................................... 13 Figure 3: Percentage of persons aged 60 and over in 2012 and 2050 ........................................ 14 Figure 4: Development of age groups 1950 to 2100 in more / less developed regions. ............ 15 Figure 5: ICF classification incorporating environmental and personal factors, WHO 2011. .. 17 Figure 6: Disability prevalence rates for thresholds 40 and 50 derived from multi-domain functioning levels in 59 countries, by country income level, sex, age, place of residence, and wealth, WHO 2011...................................................................................................................... 19 Figure 7: Age-specific disability prevalence, derived from multidomain functioning levels in 59 countries, by country income level and sex, WHO 2011. .......................................................... 20 Figure 8: Historical development of the concepts work and technology. .................................. 25 Figure 9: HAAT-model by Cook & Hussey, 1995. .................................................................... 28 Figure 10: Different problems related to working memory. ...................................................... 29 Figure 11: Mental states resulting from the interaction between level of challenge and skill. .. 31 Figure 12: Consoles Wii with Wii Remote (left) and X-Box 360 with Kinect (right). ............. 34 Figure 13: Virtual continuum according to Nilsson & Johansson 2008. ................................... 37 Figure 14: Three systems of MTM based on various levels of detail. ....................................... 41 Figure 15: Industrial assembly table constructed by the Schnaithmann GmbH. ....................... 43 Figure 16: Tools and processes for the Digital Factory and the Virtual Factory. ...................... 45 Figure 17: The software Process Designer within Siemens Tecnomatix allows to create detailed simulations of production processes. .......................................................................................... 47 Figure 18: The Microsoft Kinect (bottom) and the Asus Xtion (top). ....................................... 48 Figure 19: Softkinetic DepthSense 311. ..................................................................................... 49 Figure 20: Projected assessment game based on the UbiDisplays framework. ......................... 52 Figure 21: Wii with Wii Remote (left), Wii Remote with Wii Motion Plus (right). ................. 54 Figure 22: A senior citizen using an exergame developed for balance training. ....................... 56 Figure 23: An implementation of the Poka Yoke concept using an impression. ....................... 57 Figure 24: The Quality Assist System based on ultrasonic waves. ............................................ 59 Figure 25: Pick to light display "PickTerm Flexible". ............................................................... 60 Figure 26: Projection of visual data on welding sports. ............................................................. 61 Figure 27: An assistive system in the medical domain using projected interfaces. ................... 62 Figure 28: MEESTAR-model by Manzeschke et al. 2013. ........................................................ 65 List of Figures ix Figure 29: Number of processes at a single workplace. .............................................................75 Figure 30: Requested features of a new assistive system. ..........................................................76 Figure 31: Adapted HAAT-model. .............................................................................................79 Figure 32: Potential framework for an adaptive game system by Charles et al. ........................ 80 Figure 33: Flow curves ................................................................................................................ 82 Figure 34: CAAS-Model – abstract............................................................................................. 83 Figure 35: CAAS-model – detailed............................................................................................. 84 Figure 36: System overview. ....................................................................................................... 87 Figure 37: Experimental assembly table with motion recognition and projection..................... 88 Figure 38: Illustration of Adjoined Spheres Approach. .............................................................. 90 Figure 39: Target area states: Occluded (red), neutral (blue), occupied (green). ....................... 91 Figure 40: Detecting three-dimensional object orientation with interactive areas. .................... 92 Figure 41: Architecture of the ASED software. ..........................................................................94 Figure 42: Instructions on a monitor (left) and as a projection (right). ......................................95 Figure 43: Class hierarchy of designer elements in ASED......................................................... 96 Figure 44: Class hierarchy of panels used in ASED. .................................................................. 97 Figure 45: Screenshot of the ASED designer view..................................................................... 98 Figure 46: Screenshot: Configuring a 3D trigger area. ............................................................... 99 Figure 47: ASED Designer with maximized “Augmented Camera Preview”. .......................... 99 Figure 48: Screenshot of the ASED Runner surveillance monitor. ..........................................101 Figure 49: Design study of the gamification element (left) and final version (right). .............104 Figure 50: Screenshot of the instruction (left) and the gamification component (right) with shadow brick.............................................................................................................................. 106 Figure 51: The gamification component’s visual feedback makes use of smileys. .................108 Figure 52: Example: Process durations of one user in eight sequences. ..................................108 Figure 53: Lego objects with various degrees of complexity in the pre-study......................... 111 Figure 54: Histogram of the three test populations. ..................................................................112 Figure 55: Test populations’ mean wage-level (left) and age (right) with SD. ........................113 Figure 56: Instructions in the scenarios Gamification (left) and Projection (right). ................117 Figure 57: Logfile Analyzer list view before optimization.......................................................120 Figure 58: Logfile Analyzer list view after optimization. ........................................................121 Figure 59: Undercarriage assembled correctly (left) and with 100% error rate (right). ...........122 Figure 60: Sequence completion time of the SotA population. ................................................124 x List of Figures Figure 61: Error rate per sequence of the SotA population. ..................................................... 124 Figure 62: Sequence completion time of the Projection population. ....................................... 125 Figure 63: Error rate per sequence of the Projection population. ............................................. 125 Figure 64: Sequence completion time of the Gamification population.................................... 126 Figure 65: Error rate per sequence of the Gamification population......................................... 126 Figure 66: Histogram of average assembly time (eight sequences). ........................................ 127 Figure 67: Histogram of average error rate (eight sequences). ................................................ 127 Figure 68: Mean sequence durations with SD of the test populations. .................................... 129 Figure 69: Development of mean sequence completion times over eight sequences. ............. 130 Figure 70: Development of mean process completion times. .................................................. 131 Figure 71: Mean error rates with SD of the test populations.................................................... 132 Figure 72: Development of mean sequence error rates. ........................................................... 133 Figure 73: Development of mean process error rates. .............................................................. 134 Figure 74: Development of mean error rates of process 6 in eight sequences. ........................ 134 Figure 75: Production times (left) and errors (right) in the complete population (left column) and the faster group (right column). ................................................................................................ 135 Figure 76: Temporal development of mean error rates in the subgroups................................. 136 Figure 77: User checking the instruction and the visual gamification element. ...................... 149 Figure 78: User comparing the assembled product with the 1:1 model. .................................. 156 Figure 79: A future setup with multiple sensors combining motion and object recognition... 164 Figure 80: Detailed process times of the SotA population: graph............................................ 175 Figure 81: Detailed process times of the Projection population: graph. .................................. 177 Figure 82: Detailed process times of the Gamification population: graph............................... 179 List of Tables xi LIST OF TABLES Table 1: Thesis chapter structure................................................................................................... 5 Table 2: Summary of research questions. .....................................................................................7 Table 3: Example of MTM analysis for basic movements. ........................................................41 Table 4: Comparison of motion detection sensors available early 2012: ...................................50 Table 5: Requirements for an ideal CAAS.................................................................................. 73 Table 6: Conditions for flow and corresponding design approach. ..........................................105 Table 7: Feedbacks of the gamification component .................................................................106 Table 8: Scenario-specific questions posed after the assembly phase. .....................................116 Table 9: Instructions for the eight assembly steps. ................................................................... 118 Table 10: Means and standard deviations of sequence durations and errors............................ 128 Table 11: Questionnaire item 1: General condition ..................................................................137 Table 12: Questionnaire item 2: Nervousness ..........................................................................137 Table 13: Questionnaire item 3: Anticipation ...........................................................................138 Table 14: Questionnaire item 4: Computer experience ............................................................138 Table 15: Questionnaire item 5: Experience with computer or console games .......................138 Table 16: Questionnaire items 6 and 10: Difficulty ..................................................................139 Table 17: Questionnaire items 7 and 8: Acceptance .................................................................140 Table 18: Questionnaire items 9 and 15: Usability: Ease of Handling .....................................140 Table 19: Questionnaire items 13 and 16: Usability: Learning ................................................141 Table 20: Questionnaire items 11 and 12: Generic User Interface ...........................................142 Table 21: Questionnaire item 14: Self-Perception and Strain ..................................................143 Table 22: Questionnaire items 17, 18 and 19: Instructions.......................................................144 Table 23: Questionnaire item 20 (Projection): Instruction .......................................................145 Table 24: Questionnaire items 21 and 22 (Projection): Additional Product Model .................146 Table 25: Questionnaire items 20, 21 and 24 (Gamification): Game .......................................146 Table 26: Questionnaire items 22 and 23 (Gamification): Comments .....................................147 Table 27: RQ 1: Requirements of CAAS.................................................................................. 153 Table 28: Generic model for CAAS.......................................................................................... 154 Table 29: Implementation of motion recognition .....................................................................155 Table 30: Implementation of projection.................................................................................... 156 Table 31: Implementation of gamification................................................................................ 157 xii List of Equations, Code and Data Table 32: Quantitative impact of augmentations on speed and quality.................................... 158 Table 33: Quantitative impact on users: ................................................................................... 159 Table 34: Qualitative impact on users: ..................................................................................... 160 Table 35: Ethical dimension of CAAS: .................................................................................... 160 Table 36: Pre-experiment questions – identical in all scenarios. ............................................. 168 Table 37: Post-experiment questions: SotA.............................................................................. 169 Table 38: Post-experiment answers: SotA. ............................................................................... 170 Table 39: Post-experiment questions: Projection. .................................................................... 171 Table 40: Post-experiment answers: Projection........................................................................ 172 Table 41: Post-experiment questions: Gamification................................................................. 173 Table 42: Post-experiment answers: Gamification. .................................................................. 174 Table 43: Detailed process times of the SotA population: values. ........................................... 176 Table 44: Detailed process times of the Projection population: values.................................... 178 Table 45: Detailed process times of the Gamification population: values. .............................. 180 LIST OF EQUATIONS, CODE AND DATA Equation 1: Z-values in sphere.................................................................................................... 91 Equation 2: Comparison with reference values. ......................................................................... 91 Equation 3: Transformation of distorted projection using CMAT. ............................................ 96 Equation 4: Computation of the distances to the center of a trigger box. ................................ 101 Equation 5: Log entry generated by ASED Runner. ................................................................ 102 Equation 6: Process Time assessment in the gamification component. ................................... 107 Equation 7: Log entry generated by ASED Runner. ................................................................ 119 Equation 8: Video frame grabber function for RGB data......................................................... 166 Equation 9: OpenNI and NITE configurator function for depth data. ..................................... 167 Equation 10: Setup of the trigger box. ...................................................................................... 167 Equation 11: Check of the trigger box. ..................................................................................... 168 List of Abbreviations LIST OF ABBREVIATIONS English translations in [square brackets] AAL: Ambient Assisted Living ACM: Association of Computing Machinery AI: Artificial Intelligence AIR: Adobe Integrated Runtime ANOVA: Analysis of Variance API: Application Programming Interface AR: Augmented Reality ASLM: Assistenzsysteme für leistungsgeminderte Menschen in der Montage [Assistive sys tems for persons with impairments working in assembly] ASED: Assistive System Experiment Designer BWH: Beschützende Werkstätte Heilbronn [Sheltered Work Organization Heilbronn] CAD: Computer Aided Design CAM: Computer Aided Manufacturing CAP: Computer Aided Planning CAAS: Context-aware Assistive System CMAT: Concordance-Based Medial Axis Transform CNC: Computerized Numerical Control CPS: Cyber-Physical Systems DIN: Deutsches Institut für Normung [German Institute for Standardization] ERP: Enterprise Resource Planning EU: European Union FACS: Facial Action Coding System GOMS: Goals, Operators, Methods and Selection (rules) GPS: Global Positioning System HCI: Human-Computer Interaction HE: Hochschule Esslingen [University of Applied Sciences Esslingen] HUD: Head-up Display HMD: Head-Mounted Displays HMI: Human-Machine Interaction HTN: Hierarchical Task Network (Planning) ICF: International Classification of Functioning, Disability and Health IEEE: Institute of Electrical and Electronics Engineers xiii xiv List of Abbreviations ISO: International Organization for Standardization MEESTAR: Model for the Ethical Evaluation of Socio-Technological Arrangements MR: Mixed Reality MTM: Methods-Time Measurement NC: Numerical Control NI: Natural Interaction NITE: Natural Interaction Technology for End-user OpenCV: Open Source Computer Vision Library OpenNI: Open Natural Interaction PHP: Hypertext Preprocessor, formerly Personal Home Page Tools PLC: Programmable Logic Controller PLM: Product-Lifecycle-Management POCL: Partial-Order Causal Link (Planning) PPS: Production Planning and Steering R: Requirement REFA: Verband für Arbeitsgestaltung, Betriebsorganisation und Unternehmensentwicklung [Association for Ergonomics, Company Organization and Corporate Development] RFID: Radio-Frequency Identification RQ: Research question RGB: Red Green Blue SD: Standard Deviation SotA: State-of-the-art SUS: System Usability Scale TMU: Time Measurement Unit ToF: Time-of-Flight UN: United Nations UX: User Experience VDI: Verein Deutscher Ingenieure [Association of German Engineers] VIS: Institute for Visualization and Interactive Systems (University of Stuttgart) VR: Virtual Reality WHO: World Health Organization Introduction: Motivation 1 1 INTRODUCTION 1.1 Motivation Technology advances must not increase the gap between people who are digital im migrants, who are older or suffer from impairments1 and those who are not. Some times these advances open radically new perspectives and even lead to the “full inclusion of [these] individuals […] in the mainstream of society” (Cook, 2010). In the case of assistive technology, systems which are able to consider both the context and the user in real-time would offer better assistance, increase productivity, make work more secure and potentially even more enjoyable. In the form of route guidance systems, context-aware assistive systems (CAAS)2 have become ubiquitous in cars and smartphones. Thus travelling has ultimately become easier and more accessible. In work environments, however, context-aware assistance focusing on the worker remains the exception. While the quality gates established in modern production lines succeed in removing “waste” (i.e. failed products) from the workflow, they operate in a spatial and temporal distance from the workplace and the worker. Thus workers have to rely on their skills and expertise to make the right choices and movements and lack the opportunity to learn from problems on the fly and to improve their work habits. By establishing a feedback system close to the worker, context-aware assistive sys tems in production environments potentially improve learning, increase productivity and even enhance motivation. In the following subchapters the need for such systems will be explained in more detail. 1 In concordance with the International Classification of Functioning, Disability and Health (ICF) the term “impairment” prospectively is used in a general sense including disabilities and problems resulting from old age. 2 Assistive systems sometimes are called “wizards” because they assist users in work processes, much like the broom in Goethe’s poem Sorcerer’s Apprentice (Zauberlehr ling). In this work the terms “assistive systems” and “wizards” are used synony mously. 2 Introduction: Motivation 1.1.1 Automation versus Manual Assembly Increasing efficiency is the central goal when optimizing production. For many com panies being competitive implies reducing production costs and this is linked to two major factors: the workers’ skill level and the cost for automation. However, manual work is generally slower and prone to human errors – and some technology enthusiasts already envision a world where most repetitive tasks are taken over by AI systems and robots (Ford, 2013). So why should it not be possible to au tomate production to an extent where human errors are rendered impossible? As a general rule, large lot sizes of simple tasks are indeed taken over by automation, whereas for smaller lot sizes manual assembly is often the more economical option, since the costs of automation are high. This relation is illustrated in Figure 1: Degree of automation “Sweet spot” for CAAS in production Economic advantage of automation Trend Lot size Figure 1: Relation between automation, lot size and the application of CAAS. Despite the fact that we refer to our society as digital society, the demand for custom ized products has been increasing in recent years and so has the need for manual pro duction. Production methods like “build-to-order” and “design-to-order” led to an increasing number of product variants (Kluge, 2011). A further implication is the trend towards smaller lot sizes which encourages manual production. In some domains we can even witness the return of “manufactories”. Another issue is a direct result of globalization. Many countries established high cus tom charges for fully assembled products; so to remain profitable, many high-tech manufacturers were forced to create sets of parts for components instead of selling the complete product. These sets are sent to the buyers’ countries and assembled there. Introduction: Motivation 3 The degree of assembly economical for the manufacturers varies from country to country depending on the specific custom regulations. Since these regulations are sub ject to frequent changes, export-oriented companies like Daimler or Porsche estab lished complete departments specialized on analyzing custom regulations and determining the adequate degree of assembly for various product components. Sheltered work organizations are ideally suited for packing as well as pre-assembling these components to fit the requirements of various countries. As an example 1.000 products delivered in 30 parts each might result in ten deliveries to ten countries with ten different degrees of assembly reaching from 30 parts to two parts each. These developments make wizards for production work more important because the number of applications in their “sweet spot” grows, i.e. small lot sizes and a low level of automation (see Figure 1). 1.1.2 Increased Demand for Impaired Workers The market for production work with persons with impairments becomes increasingly important. This is mainly due to three reasons: Many industrial countries oblige companies to employ a certain percentage of disabled and impaired employees, e.g. the 1990 Americans with Disabil ities Act (DeLeire, 2000) prohibits job-related discrimination against people with disabilities. The legal obligations resulting from these and analogue laws can be met by offering integrated working solutions or by contracting “sheltered work” or “supported work” organizations (see chap. 2.2.4) which focus on providing adequate conditions for impaired workers. The increased cost and time requirements for transport when outsourcing production processes to geographically remote locations make lean produc tion with little storage capacity more difficult, so reasonably priced regional alternatives become more attractive. The absolute and relative number of impaired persons is increasing due to demographical reasons (see chap. 2.2 Demography and Targeted Users). Both the company departments integrating impaired workers and the sheltered work organizations are eager to establish systems empowering their employees to meet the rising customer demands and thus become more profitable (Kronberg, 2013). CAAS offer a way to document and improve skill development and empower workers with impairments to produce more complex components faster and with lower error rates. Ideally they allow some of them to move from sheltered work to supported work or to stay in regular work contexts when impairments occur – which is especially relevant for the elderly workforce. 4 Introduction: Motivation 1.1.3 Demographic Change The growing demand for CAAS also is a result of the demographic development: especially in European countries the percentage of older employees grows (see chap. 2.2.1 Elderly Persons). The prolongation of working lifetime requires many people to work in older age. The establishment of assistive systems in various areas of life is an efficient way to meet this demographic challenge and recently the need for adequate assistance has also become apparent in the workplace (Brach & Korn, 2012). While older employees often excel in knowledge and experience (Dul et al., 2012) it has long been established that they tend to suffer from a gradual reduction of short term memory (Anders, Fozard, & Lillyquist, 1972) resulting in a decrease of learning abilities (Satre, Knight, & David, 2006). One of the results is an increase of human errors in manual production tasks. Since in industrial production these errors are de tected and tracked back to the person who made them, many slightly impaired elderly workers decide to retire early – often feeling ashamed and humiliated. CAAS offer a simple and discrete way to augment the decreasing short term memory. Thus they have the potential to empower impaired workers to do more complex work or remain in regular jobs for a longer time. Thus CAAS combine economic benefits and address the demographic change by preventing an early loss of experienced and potentially highly productive workforce. While in the first step CAAS clearly are developed for elderly and impaired users, in a second step – in the sense of universal design – they are intended to be used by everyone working in production (see chap. 8.3 Future Research). 1.1.4 Pursuing the Ethical Good Although this work primarily pursues technological questions, we also have been con fronted with ethical problems concerning the relation between technology, work and human beings. Since CAAS need to “come close” to their users to gather sensible data, their design should be subject to ethical standards. CAAS are being developed and will be developed in the future because current sensor and computer technology allow it. It is our intention to integrate ethical aspects into this development and thus contribute to a future state of the art incorporating ethical and technological concepts. We do so by scaling the assistance according to the user’s individual physical and emotional needs and by implementing motivating elements that counter the disembodiment of work (Cardoso Machado, 2011). After all, we highly appreciate the idea of creating motivation and having fun while working. 5 Introduction: Thesis Outline 1.2 Thesis Outline This thesis consists of 11 chapters. The following table uses abbreviations and sim plifications but gives an overview of the work’s structure on levels 1 and 2. It also shows which chapters are more elaborate. Table 1: Thesis chapter structure. 1. Introduction 2. Background 1.1 Motivation 1.2 Thesis Outline 1.5 Contribu tions 1.6 Publications Included 2.1 Related Dis ciplines 2.2 Demography 2.3 Ethical & Users Dimensions 2.5 Computer Science 2.6 Engineering 3. Related Work 3.1 Motion Recognition 1.3 Research Questions 1.4 Methodology 2.4 Psychology 3.2 Projection 3.3 Gamification 3.4 Assistive Systems at Work 3.5 Ethical Standards 4. Requirements 4.1 Standards 4.2 Constraints 4.3 Study 5. Model 5.1 Adapted HAAT 5.2 Adaptive Game System 5.3 Flow Curves 5.4 CAAS Model 6. Implementa tion 6.1 Overview 6.2 Physical Integration 6.3 Motion Recognition 6.4 Instructions & Performance 7.1 Participants 7.2 Procedure 7.3 Apparatus 7.4 Data 7.5 Experiment Results 7.6 Question naire Results 7.7 Qualitative Findings 7.8 Discussion 8. Conclusion 8.1 Implications on Work 8.2 Ethical Implications 8.3 Future Research 9. Supplement 9.1 Implementa tion Details 9.2 Question naire Details 9.3 Study Result Details 6.5 Gamification 7. Studies & Evaluation 10. References 11. Index 6 Introduction: Thesis Outline The introduction chapter describes the motivation for this work, outlines the research questions and methodology and lists the major research contributions. The background chapter contains an overview of the relevant concepts and methods from ethics, psychology, computer science and engineering. The related works chap ter adds more detailed information about state-of-the-art technical implementations (e.g. motion recognition and projection) or describes models and methods directly used in the CAAS framework (gamification, assistive systems and ethical standards). The requirements chapter describes how the requirements were established and pre sents a market study. It is followed by the model which addresses the established requirements by proposing an ideal model for CAAS. In chapter 6 we describe an exemplary implementation of the CAAS covering essen tial aspects of the requirements and the model like motion recognition, projection and gamification. The prototype is extensively evaluated in chapter 7. The conclusion sums up the findings and resumes the research questions. The supplement section contains additional material that might be of interest for sub sequent studies and implementation efforts based on this work. The reference section is finally followed by the index section (chapter 11) which allows to quickly access relevant parts of this work and check where concepts or per sons are introduced first. 7 Introduction: Research Questions 1.3 Research Questions The societal developments mentioned above constitute a need for context-aware as sistive systems (CAAS) is industrial work, i.e. systems that provide cognitive assis tance for impaired and elderly workers. The aim of the research presented here is to specify the requirements for such systems, prescind a generic model, describe the implementation of a prototypical augmented system for production workplaces and evaluate the effects these augmented systems have on work and workers. An additional focus are the ethical implications of CAAS. Table 2: Summary of research questions. No. Research Question (RQ) Chapters RQ1 Which requirements should CAAS in production environments address? 4, 8 RQ2 How can the requirements be prescinded to a generic model for CAAS? 5, 8 RQ3 How can motion recognition be implemented in CAAS? 6.3, 6.4, 8 RQ4 How can projection be implemented in CAAS? 6.4.1, 7.2.2, 8 RQ5 How can gamification be implemented in CAAS? 6.5, 8 RQ6 What is the quantitative impact on work speed and quality when work is augmented by projection and gamification? 7.5, 8 RQ7 What is the quantitative impact on users when work is augmented by projection and gamification? 7.6, 8 RQ8 What is the qualitative impact on users when work is augmented by projection and gamification? 7.7, 8 RQ9 What is the ethical dimension of CAAS and how can they be designed to be humane? 2.3, 8.2 8 Introduction: Methodology 1.4 Methodology CAAS augmenting the work directly at the workplace are a new field of research. While a large part of this work is based on technologies which have been available for decades (assistive technology using computers, projector systems and computer game design), a decisive element has only recently emerged: motion recognition with 3D body tracking which is applicable outside of research labs. Affordable marker-less human body tracking was a premise for research on process-oriented assistance at workplaces. It allows the kind of implicit and natural interaction (NI) required for a low-key assistance which is “always on”. Thus the development of CAAS initially was technology-driven. The lack of adequate reference systems in the production domain led to a bottom-up approach where the requirements shaped the model, the model determined the imple mentation and the evaluation measured the system’s effects on work and workers. While this practical approach was extremely useful to identify and address fundamen tal challenges, several iterations of this process are needed to create a market-ready CAAS addressing all requirements. 1.4.1 State-of-the-Art and Requirement Studies The first approach was to look at the stakeholders in domain: providers of assistive systems for work (see chap. 3.4) and production companies potentially using such systems. These preliminary studies were necessary to establish the state-of-the-art of assistive systems in production and to identify the delta between this state and the technological possibilities as well as the desires of the production companies. Many under-graduates contributed in this process. Of particular value was the cooperation with the colleagues from production management. The derived requirements are de scribed in chapter 4. 1.4.2 Prototypes The only way of finding out how the augmentations described in the requirements affect work and workers, was building a prototype and testing it in the field. This was accomplished within the project ASLM (Assistive Systems for Persons with Impair ments Working in Assembly) and subsequently in the project motionEAP (System for Increased Efficiency and Assistance in Production Processes in Factories based on Motion Detection and Projection). Thus many students as well as the engineering company Schnaithmann contributed to the prototypes’ development (see Acknowledgements). The implementation details are described extensively in chapter 6. Introduction: Research Contributions 9 1.4.3 Evaluation Evaluating the behavior of workers with a new prototype is problematic because production is characterized by repetitions and standardized patterns of behavior. However it was considered an organizational and financial impossibility to get each worker accommodated with the prototype over several days before testing his or her “normal” behavior. The solution to this problem was to additionally map the state-of-the art in the prototype and thus “standardize” the alienating effect of the new device over all groups and scenarios. Since the tested workers at the sheltered work organization did not use workplaces with displays before, the state-of-the-art was a new experience for them. The evaluation benefited from the support of several undergraduates and the expertise of the works managers and supervisors at the Beschützende Werkstätte Heilbronn (see Acknowledgements). 1.5 Research Contributions This thesis makes several contributions to the field of human-computer interaction (HCI) with a focus on assistive technologies: we identify and describe the requirements for future context-aware assistive systems (CAAS) at the workplace we provide a generic model for CAAS we describe the implementation of a prototype we evaluate the prototype and identify future research areas we analyze and describe the qualitative and quantitative impact of the prototype’s two augmentations projection and gamification we discuss the ethical dimension of CAAS and outline a humane approach for the establishment of CAAS at workplaces 1.6 Publications The work presented here is based on the following publications (in chronological order, starting with recent publications): Korn, O., Funk, M., Abele, S., Hörz, T. & Schmidt, A. (2014) Contextaware Assistive Systems at the Workplace. Analyzing the Effects of Projection and Gamification. In PETRA ’14 Proceedings of the 7th International Conference on PErvasive Technologies Related to Assistive Environments. New York, NY, USA: ACM. [in print] 10 Introduction: Publications Funk, M., Korn, O., & Schmidt, A. (2014). An Augmented Workplace for Enabling User-Defined Tangibles. In CHI EA ’14 Extended Abstracts of the ACM SIGCHI Conference on Human Factors in Computing Systems (pp. 1285-1290). New York, NY, USA: ACM. Funk, M., Korn, O., & Schmidt, A. (2014). Assisitive Augmentation at the Manual Assembly Workplace using In-Situ Projection. In CHI ’14 Workshop on Assistive Augmentation. April 27th 2014. Korn, O., Abele, S., Schmidt, A., Hörz, T. (2013): Augmentierte Produktion. Assistenzsysteme mit Projektion und Gamification für leistungsgeminderte und leistungsgewandelte Menschen. In: Boll, S.; Maaß, S. & Malaka, R. (eds.): Tagungsband der Konferenz Mensch & Computer 2013 (pp. 119– 128), Oldenbourg Verlag, Munich. Korn, O., Schmidt, A., & Hörz, T. (2013). Augmented Manufacturing: A Study with Impaired Persons on Assistive Systems Using In-Situ Projection. In PETRA ’13 Proceedings of the 5th International Conference on PErvasive Technologies Related to Assistive Environments (pp. 21:1–21:8). New York, NY, USA: ACM. doi:10.1145/2504335.2504356 Korn, O., Schmidt, A., & Hörz, T. (2013). The Potentials of In-Situ-Projection for Augmented Workplaces in Production. A Study with Impaired Persons. In CHI EA ’13 Extended Abstracts of the ACM SIGCHI Conference on Human Factors in Computing Systems (pp. 979–984). New York, NY, USA: ACM. doi:10.1145/2468356.2468531 Korn, O., Brach, M., Schmidt, A., Hörz, T., & Konrad, R. (2012). ContextSensitive User-Centered Scalability: An Introduction Focusing on Exergames and Assistive Systems in Work Contexts. In S. Göbel, W. Müller, B. Urban, & J. Wiemeyer (eds.), E-Learning and Games for Training, Education, Health and Sports (Vol. 7516, pp. 164–176). Berlin, Heidelberg: Springer Berlin Heidelberg. Retrieved from http://www.springerlink.com/index/10.1007/978-3-642-33466-5_19 Korn, O., Schmidt, A., Hörz, T., & Kaupp, D. (2012). Assistive system experiment designer ASED: A Toolkit for the Quantitative Evaluation of Enhanced Assistive Systems for Impaired Persons in Production. In ASSETS ’12 Proceedings of the 14th international ACM SIGACCESS conference on Computers and Accessibility (pp. 259–260). Presented at the ASSETS ’12, Boulder, Colorado, USA: ACM Press. doi:10.1145/2384916.2384982 Introduction: Publications 11 Korn, O. (2012). Industrial playgrounds: how gamification helps to enrich work for elderly or impaired persons in production. In EICS ’12 Proceed ings of the 4th ACM SIGCHI symposium on Engineering Interactive Com puting Systems (pp. 313–316). New York, NY, USA: ACM. doi:10.1145/2305484.2305539 Korn, O., Schmidt, A., & Hörz, T. (2012). Assistive systems in production environments: exploring motion recognition and gamification. In PETRA ’12 Proceedings of the 5th International Conference on PErvasive Technol ogies Related to Assistive Environments (pp. 9:1–9:5). New York, NY, USA: ACM. doi:10.1145/2413097.2413109 12 Background: Related Disciplines 2 BACKGROUND Since its foundation in the nineties as described in Assistive technologies: principles and practice (Cook & Hussey, 1995), assistive technology has always been an inter disciplinary field. While it originally focused on persons with motoric or cognitive impairments, assistive technology has long transcended these traditional boundaries. Route guidance systems in cars have become regular aids in everyday life and more recent developments like Google Glass show that the desire for “enhancements” or “augmentations” of the human body is not only a technical possibility but marketdriven. In spite of these rapid advances Vanderheiden comes to a skeptical conclusion on the future of assistive technologies in an article on “ubiquitous accessibility”: However, the high cost of assistive technologies, especially assistive technologies that can cope with ever-evolving mainstream technologies, is putting the assistive technology that will work with newer mainstream technologies out of the reach of many or most. (Vanderheiden, 2008) The following sub-chapters will show that this view has to be revised: the “ever-evolv ing mainstream technologies” partly diffuse into the realm of assistive technologies allowing new and especially comparatively low-price applications. The combination of low-price motion recognition with low-price projectors as described and evaluated in this work is a good example of this development. In fact, these advances allow assistive technologies with capabilities so far beyond established forms of assistance that the ethical implications of their usage have to be carefully considered (see chap ters 2.3 Ethical Dimensions, 8.2 Ethical Implications: Towards Humane Work). 2.1 Related Disciplines To make this work more accessible to readers from various disciplines, we will briefly outline relevant concepts for the design and development of context-aware assistive systems (CAAS) in the following sub-chapters: Engineering: production management, digital factories and cyber-physical systems (CPS), methods-time measurement (MTM), assembly workplaces and as sistive systems in production environments Computer Science: human-computer interaction (HCI), assistive technologies, user-centered software design or user experience design (UX), augmented reality (AR), 13 Background: Related Disciplines motion recognition, implicit interaction, natural interaction (NI) and artifi cial intelligence (AI) Psychology and Ethics: assisted working, cognitive workloads, flow and motivation, ethical issues Engineering Production Management Digital Factory Assembly Workplaces CPS Context-Aware Assistive Systems AR HCI MTM Assistive Technologies NI Implicit Interaction Motion Recognition Computer Science UX AI Ethical Issues Flow Cognitive Workloads Assisted Working Psychology & Ethics Figure 2: Disciplines related to context-aware assistive systems. Although these disciplines partly overlap and finally converge in the research topic presented here, they are vast research areas. For this reason the background and the state-of-the art chapters focus only on areas and findings relevant for the research on context-aware assistive systems presented here. 14 Background: Demography and Targeted Users 2.2 Demography and Targeted Users 2.2.1 Elderly Persons According to the United Nations World Population Prospects (United Nations, De partment of Economic and Social Affairs, Population Division, 2013) even on a global scale the population aged 60 or over (“the elderly”) is the fastest growing. In the more developed regions, the ratio of the elderly is increasing at 1.0% per year before 2050 and 0.11% annually from 2050 to 2100; it is expected to increase by 45% by the mid dle of the century, rising from 287 million in 2013 to 417 million in 2050 and to 440 million in 2100. In the less developed regions, the population aged 60 or over is currently increasing at the fastest pace ever, 3.7% annually in the period 2010-2015 and is projected to increase by 2.9% annually before 2050 and 0.9 per cent annually from 2050 to 2100; its numbers are expected to rise from 554 million in 2013 to 1.6 billion in 2050 and to 2.5 billion in 2100. Figure 3: Percentage of persons aged 60 and over in 2012 and 2050 Adapted from United Nations, Department of Economic and Social Affairs, Population Division, Population Ageing and Development 2012 Background: Demography and Targeted Users 15 The UN World Population Prospects also indicates reasons for this development: increased life expectancy baby boom generation reaching the sixties inversion in the population pyramid concentrated around sixties Accordingly, one of the most important challenges governments around the world are facing is maintaining or improving the quality of life for the elderly. This is especially relevant for the “more developed regions”3 where the proportion of the cohort aged 60 and over is growing the fastest: Figure 4: Development of age groups 1950 to 2100 in more / less developed re gions. Adapted from United Nations, Department of Economic and Social Affairs, Population Division, Population Ageing and Development 2012. As Figure 4 shows, the population graphs move almost in parallel in the less devel oped regions while in the more developed regions the 60+ cohort is growing the fastest and from 2025 onwards will represent most persons. The rapid growth of this cohort is referred to as “demographic change” by the affected countries; it will continue until about 2060 when it will transform into a sideward movement with only slight growth. 3 In UN terminology the more developed regions comprise all regions of Europe and Northern America, as well as Australia, Japan and New Zealand. The less developed regions comprise all regions of Africa, Asia (excluding Japan) and Latin America and the Caribbean, as well as the regions of Melanesia, Micronesia and Polynesia. 16 Background: Demography and Targeted Users In the next 50 years the more developed regions will have to find technological, social and ethical solutions to the ever-growing relative and absolute numbers of elderly people. A good measurement is the “old-age support ratio”: the number of persons aged 15-64 per person aged 65 or over. These working-age persons are the potential caretakers and also the potential taxpayers supporting social and medical institutions used by the elderly. In Germany, Italy, Japan and Sweden there are only 3 workingage individuals for each older person. The only country with less young persons per elderly person (about 2.5) is Japan. By contrast, countries such as Bahrain or the United Arab Emirates have over 30 persons of working age per older person. Euro pean countries tend to cluster at the lower end of support ratios, while countries from Western Asia, South-Central Asia and sub-Saharan Africa tend to be at the higher end. Most countries in the world are in an intermediate transitional phase, with old-age support ratios between 5 and 20 persons of working-age per older person. Obviously the European Union (EU) is the first cluster of nations to experience and to address the challenges of a rapidly increasing ratio of elderly people. Accordingly the “Ambient Assisted Living Joint Programme” (AAL) has been installed both on the European level and on the national levels of the 22 participating European coun tries. The AAL-programme tries to bring together the special user requirements of elderly persons and technological solutions. After focusing on health, care and as sisted homes in 2013 with the 6th call “ICT-based Solutions for Supporting Occupa tion in Life of Older Adults” the research focus broadened to integrate the domain of work in an office, a factory or any working environment […] and to strengthen the industrial base in Europe” (AAL Contents Working Group, Task Force, 2013). 2.2.2 Disabled and Impaired Persons When we talk about “the elderly” it has been established that the term refers to persons aged 60 and above. However, when we talk about “disabilities” the classification is far more difficult – especially since recent approaches aim to integrate the interaction of disabled individuals with the society they live in. In the International Classification of Functioning, Disability and Health (ICF) the WHO (World Health Organization) defines disabilities as follows: Disability is an umbrella term for impairments, activity limitations and par ticipation restrictions. It denotes the negative aspects of the interaction be tween an individual (with a health condition) and that individual’s contextual factors (environmental and personal factors). (World Health Organisation, 2001) Background: Demography and Targeted Users 17 The ICF categorized problems with human functioning in three interconnected areas: impairments are problems in body function or alterations in body structure, e.g. paralysis or blindness; activity limitations are difficulties in executing activities, e.g. walking or eating; participation restrictions are problems with involvement in any area of life, e.g. facing discrimination in employment or transportation The advantage of this classification is the widespread acceptance of the ICF. It was officially endorsed by all 191 WHO Member States in the 54th World Health Assem bly in 2001. It marks a decisive shift in the understanding of the concept of disability best described by the ICF itself: The ICF puts the notions of ‘health’ and ‘disability’ in a new light. It acknowledges that every human being can experience a decrement in health and thereby experience some degree of disability. Disability is not some thing that only happens to a minority of humanity. The ICF thus ‘main streams’ the experience of disability and recognises it as a universal human experience. By shifting the focus from cause to impact it places all health conditions on an equal footing allowing them to be compared using a com mon metric – the ruler of health and disability. Furthermore ICF takes into account the social aspects of disability and does not see disability only as a 'medical' or 'biological' dysfunction. By including Contextual Factors, in which environmental factors are listed ICF allows to records the impact of the environment on the person's functioning. (World Health Organisation, 2001) Disability now refers to the negative aspects of the interaction between individuals with a health condition (such as cerebral palsy, Down syndrome, depression) and per sonal and environmental factors (such as negative attitudes, inaccessible transporta tion and public buildings, and limited social supports). The following models shows how a combination internal and external factors affects activities: Figure 5: ICF classification incorporating environmental and personal factors, WHO 2011. 18 Background: Demography and Targeted Users In many countries this holistic approach or “bio-psycho-social model” has not yet been translated into national law. The authoritative German Code of Social Law IX (Sozialgesetzbuch IX) defines disabilities as follows: Human beings are disabled if their bodily function, their mental abilities or their emotional health with a high probability deviate from the typical state in their age for more than six months and for that reason their participation in societal life is compromised. 4 (SGB IX, 2001, § 2.1) In this text, disability is still described as an impairment of the individual which might have societal consequences – whereas the ICF’s definition allows an impairment in itself to be a result of societal behavior, e. g. mental and physical barriers. Germany’s Code of Law is no exception. In most countries the ICF’s universal understanding of disabilities is just beginning to be established both in the existing medical frameworks and in the organizations working with impaired persons. Accordingly the harmonization and standardization of question sets for assessment of health and disability at the level of different nations are in progress but far from being established. The data presented here are based on the World Health Survey (World Health Organization, 2004) conducted 2002 to 2004 which is “the largest multina tional health and disability survey ever using a single set of questions and consistent methods to collect comparable health data across countries” (World Health Organisa tion & World Bank, 2011, p. 25). The following passages are based on this survey. A total of 70 countries were surveyed of which 59, representing 64% of the world population, had weighted data sets that were used for estimating the prevalence of disability of the world’s adult population aged 18 years and older. The survey aggre gated a disability score ranging from 0 to 100, where 0 represented “no disability” and 100 “complete disability”. The average prevalence rate in the adult population aged 18 years and over derived was 15.6% (some 650 million people of the estimated 4.2 billion adults aged 18 and older in 2004) ranging from 11.8% in higher income countries to 18.0% in lower in come countries. This figure refers to adults who experienced significant functioning difficulties in their everyday lives. The average prevalence rate for adults with very significant functioning difficulties was estimated at 2.2% or about 92 million people in 2004. 4 German original: Menschen sind behindert, wenn ihre körperliche Funktion, geistige Fähigkeit oder seelische Gesundheit mit hoher Wahrscheinlichkeit länger als sechs Monate von dem für das Lebensalter typischen Zustand abweichen und daher ihre Teilhabe am Leben in der Gesellschaft beeinträchtigt ist. Background: Demography and Targeted Users 19 Figure 6: Disability prevalence rates for thresholds 40 and 50 derived from multidomain functioning levels in 59 countries, by country income level, sex, age, place of residence, and wealth, WHO 2011. The alternative Global Burden of Disease study in the 2004 update (Lopez, Mathers, Ezzati, Jamison, & Murray, 2006) comes to similar numbers: It estimates that 15.3% of the world population (some 978 million people of the estimated 6.4 billion in 2004) had “moderate or severe disability” while 2.9% or about 185 million experienced “se vere disability”. Based on the population estimates for 2010 (6.9 billion people with 5.04 billion 15 years and over and 1.86 billion under 15 years) and the 2004 disability prevalence estimates (World Health Survey and Global Burden of Disease) there were around 785 (15.6%) to 975 (19.4%) million persons 15 years and older living with disability. Of these, around 110 million (2.2%) to 190 million (3.8%) experienced significant difficulties in functioning. Including children, over a billion people (or about 15% of the world’s population) were estimated to be living with disability. 20 Background: Demography and Targeted Users 2.2.3 Aging and Disability Since disabilities are usually interpreted in relation to what is considered “regular functioning”, several studies found differences between self-reported and measured aspects (Andresen, Fitch, McLendon, & Meyers, 2000; Ikeda, Murray, & Salomon, 2009). Elderly persons frequently do not think of themselves as being disabled be cause they consider their deficits as “normal” or appropriate for their age. This also depends on the degree of support and social standards: the prevalence of disability in lower income countries among people aged 60 years and above was 43.4% compared with 29.5% in higher income countries (World Health Organisation & World Bank, 2011, p. 27). In spite of these findings the overall relationship between ageing and disability is straightforward: there is higher risk of disability at older ages, with the higher disabil ity rates among older people reflecting the accumulation of health risks across a lifespan of disease, injury, and chronic illness (Australian Institute of Health and Wel fare, 2004). The correlation between age and disability prevalence is obvious in both low- and high-income countries: Figure 7: Age-specific disability prevalence, derived from multidomain function ing levels in 59 countries, by country income level and sex, WHO 2011. Given this strong correlation between age and disability prevalence, global ageing has a major influence on disability trends: both the absolute and the relative number of persons with disabilities will grow. This growth is increased by the fact that the fast est-growing age cohort worldwide, increasing at 3.9% a year, are persons aged 80 to 89 years – and at this age the rates of disability are very high (Robine & Michel, 2004). While elderly persons are still at work, the awareness of impairments is increased: although elderly employees often excel in knowledge and experience, it has been es tablished 40 years ago that their short term memory is continually declining (Anders Background: Demography and Targeted Users 21 et al., 1972) which results in an increase of human errors in complex activities (Salt house, 1990) as well as a decrease of learning abilities (Satre et al., 2006). The effects of cognitive deficits are discussed in more detail in chap 2.4.2 Cognitive Workloads. Since cognitive deficits currently are hardly countered by technological methods5, many elderly workers decide to retire early as soon as they feel the impact on their regular functioning. In production environments, such deficits are recognized very early, because time criticality is combined with frequent quality controls (see chap. 4.2.2 Total Quality). The decision to retire early has negative impacts on the economy (a former worker and taxpayer becomes a holder of an annuity and the old-age support ratio declines) and potentially also on the individual (work is an essential component for self-esteem, see chap. 2.3.2 Value of Work). For this reason cognitive deficits should actively be countered by assistive systems which support the workers cogni tively – both in regular workplaces and in sheltered or supported work. 2.2.4 Sheltered Work and Supported Work The concept of sheltered work or supported work is offering a protected environment to persons with impairments who cannot compete on the regular job market. There are many different forms of sheltered employment reaching from dedicated institu tions to daycare centers. Virtually all forms of sheltered employment can generally be classified into two types (Kregel & Dean, 2002): Supported work: Transitional employment programs intended to provide training and experience to individuals in segregated settings, so that they will be able to acquire the skills necessary to succeed in subsequent com petitive employment. Sheltered work: Extended employment programs designed for long-term or permanent placements for individuals that will allow them to use their abil ities to earn wages in the segregated setting. The majority of adults with cognitive disabilities work in segregated, sheltered em ployment settings. Although sheltered work organizations fulfill a social task in providing sheltered working conditions, in many countries they are also required to work economically by law, e.g. in Germany (Kronberg, 2013). However, from an economic perspective it would be beneficial to empower impaired persons to move to transitional programs, because sheltered work organizations normally depend on cash 5 Instead organizational methods are used, e.g. reducing the work tasks’ complexity. This felt loss of status often is combined with a transfer into another team which again causes social stress. As a result many workers prefer to quit their job completely. 22 Background: Demography and Targeted Users benefit programs. From the perspective of inclusion, i.e. the integration of persons with functioning deficits into regular contexts of living, an empowerment of the users would also be beneficial. It has been shown that once in sheltered employment, only very few persons are able to return to competitive employment (Murphy & Rogan, 1995). A ten-year longitudinal analysis of 877 individuals with cognitive disabilities produced a significant amount of evidence indicating that integrated as opposed to sheltered employment improves the employment outcomes. The assumption that two different sets of users were compared has been falsified, as “surprisingly little differ ence exists in the demographic profiles of persons working in integrated versus shel tered employment settings” (Kregel & Dean, 2002, p. 81). Context-sensitive assistive systems (CAAS) could improve the work conditions for impaired persons in sheltered work organizations by empowering them to produce more complex products in better quality. This results in higher pay for the worker, increases the sheltered work organizations’ profits and decreases the public funding – a potential “win-win-win-situation”. CAAS also have the potential to ease the transi tion of persons from sheltered work to integrated work. A third advantage is that these assistive systems can help users to remain in regular work contexts even when impair ments occur – a perspective which is especially relevant for the elderly workforce. Background: Ethical Dimensions 23 2.3 Ethical Dimensions According to the WHO World report on disability “disability is increasingly under stood as a human rights issue.” (World Health Organisation & World Bank, 2011). This makes the use and development of assistive technology an ethical subject. To understand the value of assistive technology in work contexts and the ethical ques tions arising from its use it is helpful to look back in history to see how the concepts of work and technology changed over time. This analysis can only be very brief and is not intended for readers with an academic background in philosophy or ethics but for interested readers of other disciplines. 2.3.1 Value of Technology There was a time when science, technology and art were one – the mythological times. In ancient Greece there is a famous dialogue between the rhapsodist Ion of Ephesus and Socrates where the latter divides these three concepts (Schlaffer, 2005). He does so by proving to Ion that he may be able to sing and tell Homer’s songs, i.e. apply the art of poetry (poietike) but neither can he perform the arts or techniques (techne) de scribed in the songs himself nor does he know the concepts or the science beneath those arts (episteme). He proves this using examples like medicine, military, ship nav igation and driving chariots. Thus Socrates gives an early definition of professional ism (knowledge and ability) and separates science and technology from the fine arts and poetry, which in his understanding are not part of the world of reason. However, in ancient Greece this did not mean the arts were inferior – instead they were considered as gifts of the gods. Socrates’ view did not represent the mainstream – and many Greeks were persuaded that all important things could be learned by read ing Homer’s epic poetic works Ilias and Odyssey. In fact there is an established four fold argument of ancient technology criticism (Mitcham, 1994, p. 282): the will to technology or the technological invention often involves turning away from faith or trust in nature or providence technical affluence and the concomitant processes of change tend to under mine individual striving for excellence and societal stability technological knowledge likewise draws human beings into intercourse with the world and obscures transcendence technical objects are less real than nature With some exceptions this skeptical attitude towards technology and science perse vered for centuries through the middle ages until the age of renaissance. 24 Background: Ethical Dimensions 2.3.2 Value of Work The historical development of the value of work resembles the described development in the perception of technology. In the antiquity work was understood as a basic ne cessity best done by slaves. Even administrative work like the management of estates was considered a necessary evil. The aim was to spend time with art, religion or simply with life’s pleasures. The only exceptions were architecture, medicine, politics and military – but as the historian of art and culture Jacob Burckhardt explained these occupations were not considered work but special forms of art (Burckhardt, 1999) . This low esteem of work persevered in Christian times. In a central passage of the Bible work is described as a form of punishment: “By the sweat of your face you shall eat bread, till you return to the ground, for out of it you were taken; for you are dust, and to dust you shall return.” (Genesis 3:19, English Standard Version). Until the late medieval times the doctrine “ora et labora” (pray and work) was not only common monastic practice but also the church’s guide to a humble life agreeable to God. Work ing was by no means considered a form of self-fulfillment but a form of submission. This changed when the protestant reformer Martin Luther (1483-1546) re-interpreted the concept of “Schickung” (approximate translation: divine dispensation). He taught that salvation is not earned by good deeds but received through faith as a free gift from God. At first this seems to decrease the value of work even further. However, in the influential essay “The Protestant Ethic and the Spirit of Capitalism” (1904-1905) the German sociologist, philosopher and economist Max Weber showed that Luther dig nified mundane professions as adding to the common good and thus being blessed by God (Weber, 2009). Lutheran Protestants perceived working as devoting effort to the praise of God. In the English-speaking part of Europe a similar line of argument has been put forward by Francis Bacon (1561-1626) in maintaining that “God has given humanity a clear mandate to pursue technology [by working] as a means for the compassionate melio ration of the suffering of the human condition” (Mitcham, 1994, p. 284). This Protestant work ethic (or from a historical perspective: renaissance work ethic) was very successful. Viewing work as an end in itself or as a “calling” made work more meaningful and potentially increased work motivation. At the same time this work ethic corresponded to the capitalistic model of the world. It started to emancipate itself from the religious roots and shaped the rise of a new form of economy where a dishwasher could become a millionaire. The historical rise of the value of work and technology is illustrated in Figure 8. 25 Background: Ethical Dimensions Rise of capitalism Value of the concepts work and technology Bacon Luther Greek Antiquity Roman Antiquity Medieval Times Renaissance Today Future Figure 8: Historical development of the concepts work and technology. While Protestant work ethic raised the social prestige of work, there were strong em pirical forces pushing it downwards. The early days of industrialization where social security, working hours and security standards were largely unknown, took a high death toll. Also on the conceptual level industrialization implies a division of labor and thus a decontextualizing or as Polanyi puts it a “disembedding” of work actions (Polanyi, 1957). “Disembedded” work is driven by forces independent of the human body and often the work results also do not relate to concepts within the workers’ minds, i.e. the workers do not know what the product component they are working on is good for. Today many work conditions have changed to the better; however increasing special ization still results in a high level of disembedding in most jobs. In fact the number of critical tones regarding today’s eminent prestige of work is growing. Recent discus sions about the increasing number of “burnout” patients (Brand & HolsboerTrachsler, 2010) and the loss of “work-life balance” indicate that the prestige of work has reached a plateau and might even have come to a point where a change in trend is a likely future scenario. Nevertheless currently there still is a common social consensus that a reasonable amount of work is “good” and meaningful. By using the day reconstruction method where activities are assessed on an affective scale (Kahneman, Krueger, Schkade, Schwarz, & Stone, 2004) the positive effect of work on well-being has been shown empirically by comparing 600 employed and unemployed persons (Knabe, Rätzel, Schöb, & Weimann, 2010). Interestingly the positive effect is not derived from the time spent working – in fact the activity of working is rated lowest on the affective scale – but from the resulting income and its positive effect on the overall life satis faction (and probably the social prestige). 26 Background: Ethical Dimensions 2.3.3 Implications for the Work of Persons with Disabilities If we look at the historical development of the value of work and technology, both have reached a peak in prestige. Work has transcended the time where it was just a means to earn a living and technology has transcended the phase where it was just a tool to an end. It has already been described that within recent years there was a shift in the under standing of what being disabled means (see chap. 2.2.2 Disabled and Impaired Per sons). Instead of focusing on needs and using a “rhetoric of compassion” (Rogers & Marsden, 2013) the new focus lies on empowerment. This shift of focus implies a rising importance of the environment or the context. Designing for accessibility or designing for diversity (see chap. 2.5.2 Assistive Technology) leads to questions like: how would a work environment look like that allowed impaired persons to be in cluded? Context-aware assistive systems (CAAS) that augment regular workspaces without physically changing them clearly will be a huge step towards universal access. Within due time, implementing such systems will rise from a vision to a requirement in the industrialized countries. Already in 2008 the proactive approach of inclusion was sup ported by the United Nations (UN), when in the “Convention on the Rights of Persons with Disabilities” work was constituted a right (Article 27: Work and Employment): States Parties recognize the right of persons with disabilities to work, on an equal basis with others; this includes the right to the opportunity to gain a living by work freely chosen or accepted in a labour market and work envi ronment that is open, inclusive and accessible to persons with disabilities. States Parties shall safeguard and promote the realization of the right to work, including for those who acquire a disability during the course of em ployment, by taking appropriate steps, including through legislation, to, in ter alia […](United Nations, 2008) The idea to interpret work as a human right is a direct result of the risen prestige of the concept of work. It implies that national laws will start converting this new ap proach into binding legislation. The ethical consequences of this process are discussed in chap. 8.2 Ethical Implications: Towards Humane Work. Background: Psychology 27 2.4 Psychology When humans interact with computers, human behavior and thus psychology comes into play. In their interactions with computers humans frequently follow their emo tions and instincts. Even in their most primitive forms computers with their ascribed intelligence (see chap. 2.5.4 Artificial Intelligence) stirred emotions ranging from dis gust and bewilderment to ecstasy. Marvin Minsky’s famous 1970 statement6 that “in from three to eight years we will have a machine with the general intelligence of an average human being […] that will be able to read Shakespeare […] tell a joke, have a fight” pointedly illustrates the exuberant expectations of these early years. Over time the expectations became more rational and the awareness grew that com puters are machines that will fundamentally differ from humans for a long time to come. In the last decade, however, the advances in the fields of sensors and algorithms allowed an ever-growing part of the spectrum of human abilities to be analyzed, in terpreted and partially mirrored by computers. Thus the user experience (UX) can be designed to fit the human mindset. In the following we will introduce a model that describes the interactions between humans and assistive technology. The model is complemented by the concept of cog nitive workloads and finally by the concept of flow. Their combination offers a con ceptual framework for designing interactive systems that effect human emotions in a positive way, as described in the CAAS-model presented in this work (see chap. 5 Model and Architecture). 2.4.1 HAAT-model When human activity is assisted, behavior patterns have to be analyzed and interpreted to generate appropriate interventions. While the sensors integrated in CAAS allow to access real-world data, the selection and interpretation of this data depends on the underlying model. The Human Activity Assistive Technology Model (HAAT) (Cook & Hussey, 1995) has formalized the interactions between assistive technology and users. It is based on the basic Human Performance Model (Bailey, 1989) which already separates human, activity and context; this model is changed by adding “assistive technology” and mov ing “context” into the frame (see Figure 9). 6 The statement was published in Life Magazine. On several occasions Minsky stated that he was misquoted. Like usually this counterstatement could not balance the im pact of the alleged assertion. 28 Background: Psychology Figure 9: HAAT-model by Cook & Hussey, 1995. The HAAT-model describes four basic components and functions of assistive tech nology: activity (areas of performance, e.g. work, play, school) human (“intrinsic enabler”, including skills) context (setting, social, cultural, physical) assistive technology (“extrinsic enabler”): integrates a human-technology interface (HTI), an activity output, a processor and an environmental interface In production environments the task might be the assembly of a specific product, the user a person with cognitive impairments and the context an assembly workshop in a sheltered work organization. While the impaired person with his or her abilities and skills is seen as the intrinsic enabler, the CAAS would then be the extrinsic enabler. Interestingly the terms Cook & Hussey use to describe the human, strongly resemble the terms used to describe systems in computer science: the user can gather “sensory input”, the corresponding brain is called “central processing” (subdivided into percep tion, cognition, memory and motor control) and legs, hands and fingers etc. are called “effectors”. 29 Background: Psychology The human-technology-interface can be “unicausal or bicausal”, so a direct and natu ral interaction with the user is already inherent to the model – although the authors rather thought of prosthetic limbs than of augmented reality. The isomorphism regard ing sensor-based computational systems makes the HAAT model a good starting point for a model for CAAS (see chap. 5 Model and Architecture). 2.4.2 Cognitive Workloads Assistive systems designed for universal access also address users with impairments (see chap. 2.2 Demography and Targeted Users). With these users, some of the rec ommendations regarding cognitive support like “simple error handling” or “reducing the short-term memory load” (see chap. 4.1 Standards for Designing Interactive Sys tems) become more important than with users in best health. Salthouse analyzed the process of “cognitive aging” and found that the role of working memory is crucial (Salthouse, 1990) – a universal finding that can be generalized for disabled persons suffering from related cognitive deficits. With working memory be ing defined as the part of the brain responsible for concurrent information processing and storage, he introduces the desktop metaphor to illustrate its function: like our brains desktops are used both for processing and for storing information. Typical problems in human simultaneous processing and storage abilities can be illustrated by different “desktops”: Storage Processing Storage Storage Problem of smaller workspace Processing Storage Processing Storage Processing Regular structure of working memory Storage Problem of greater processing require ments Problem with filtering and organizing infor mation Figure 10: Different problems related to working memory. Adapted from Salthouse, 1990. Salthouse’ experiments indicated that older adults use a smaller capacity of working memory while performing cognitive tasks and that out-of-context tasks (i.e. tasks not related to everyday activities, e.g. repeating shown numbers in reversed order) are 30 Background: Psychology especially problematic. Also using new information and determining its veracity was more difficult for older people. These findings have to be taken into account when designing assistive systems with new forms of augmentations. While the augmentations can potentially support the working memory (the “storage”) they might also draw the limited processing power away from the main task. 2.4.3 Flow One of the augmentations implemented and analyzed in the context of CAAS is gam ification (see chap. 3.3). In this work gamification is seen as a means to achieve “flow” – a mental state in which a person feels fully immersed in an activity, experiences an energized focus and believes in the success of the activity. This state where high skill and adequate challenge converge was first proposed by Csíkszentmihályi in 1975. In various publications (Csíkszentmihályi & Nakamura, 2002; Csíkszentmihályi & Rathunde, 1992; Csíkszentmihályi, 1975) he and other authors identify the following factors as accompanying flow: 1. A high degree of concentration on a limited field of attention (a person en gaged in the activity will have the opportunity to focus and to delve deeply into it). 2. A loss of the feeling of self-consciousness, the merging of action and awareness. 3. Distorted sense of time, one's subjective experience of time is altered. 4. A sense of personal control over the situation or activity. 5. A lack of awareness of bodily needs (to the extent that one can reach a point of great hunger or fatigue without realizing it) 6. Absorption into the activity, narrowing of the focus of awareness down to the activity itself, action awareness merging. Background: Psychology 31 Figure 11: Mental states resulting from the interaction between level of challenge and skill. The figure above shows the spectrum of mental states and links them to challenge and skill. Thus the concept of flow inherently integrates Bailey’s human performance model described at the beginning of this chapter: high skill results in high perfor mance. However, Csíkszentmihályi adds the element of challenge and puts forward the idea, that only the combination of high skill and high challenge level can create flow. This does not imply that flow is a mental state limited to highly skilled persons. The method to achieve flow is adjusting the challenge level to meet the user’s current skill level. For this adaptation permanent feedback is required. This can either be achieved by highly interactive applications like games – or by integrating sensors that measure and interpret human behavior as described in the HAAT-model. In a further rendition of the concept, four conditions were deemed necessary to achieve the flow state (Csíkszentmihályi, Abuhamdeh, & Nakamura, 2005): One must be involved in an activity with a clear set of goals. This adds direction and structure to the task. One must have a good balance between the perceived challenges of the task and the own perceived skills. One must have confidence to be capable to do the task. 32 Background: Psychology The task at hand must have clear and immediate feedback. This helps the person negotiate changing demands and allows to adjust their performance to maintain the flow state. The activity is intrinsically rewarding, so there is a perceived effortlessness of action. Clearly achieving a flow state depends strongly on the task. Many tasks are vague and have unclear borders, e.g. writing or planning: often both time and quality are un defined so the result inevitably feels “un-finished”. If such and similar open tasks are designed to be measurable and clearly defined – like quests in games – they often feel artificial: both a word constraint for an essay and a time constraint for planning are not task-inherent. The process of re-designing real-world tasks to feel like games is called “gamification” (see chap. 3.3 Gamification). However, many tasks are princi pally highly structured and easily measurable: these tasks are principally well-suited for creating conditions that allow achieving a flow state. Amongst many others, as sembly work in production environments is such a task. Assembling countable parts provides both a clear objective and an immediate visual feedback. If tasks in production are re-designed to create and preserve a feeling of flow, they ideally have to scale to match a person’s changing level of performance. Otherwise a worker encountering a longer phase of underperformance (e.g. as a result of a physical or mental problem) would permanently receive negative feedback and quickly move to negative mental states like arousal, anxiety or worry. The required scaling of the performance intensity can be achieved by reducing the task complexity or by reducing the time requirements (see chap. 6.5 Gamification Component). Background: Computer Science 33 2.5 Computer Science 2.5.1 Human-Computer Interaction From a computer science perspective, an assistive system primarily is a computerbased system integrating data from users with special requirements. This makes as sistive systems a sub-discipline of the vast field of human-computer interaction (HCI). The communication and interaction between users and computers was highly formal ized for decades: keyboard, mouse and joystick were the only ways to manipulate things within the system. This is the time when formalized approaches like the GOMS model (Goals, Operators, Methods, and Selection rules) (Card, Moran, & Newell, 1983) or the simplified keystroke-level model (Kieras, 1993) were developed to de scribe HCI. In analyzing and describing the smallest human movements these models bear a strong resemblance to the approaches used in production management to map and plan human tasks in production processes like Methods-Time Measurement (MTM; see chap. 2.6.1 Measuring Work Processes with MTM). With the success of webcams in the mass market in the early nineties, wide-spread sensors integrated live data from the real world without the need for human interac tion. From this point onwards the domain of HCI began to grow rapidly with new forms of input (GPS, accelerometers, motion sensors) and new output devices (mobile phones, tablets, projectors) emerging in quick succession. It is not surprising that exactly in this time the idea of “ubiquitous computing” emerged. Mark Weiser, the pioneer in this field, first described the idea that the “most profound technologies are those that disappear. They weave themselves into the fabric of everyday life until they are indistinguishable from it” (Weiser, 1999, p. 3)7. He also clear-sightedly envisioned the success of “call tabs” (smart phones), pads (tablets) and predicted “hundreds of computers per room”. This idea of ubiquitous computing has soon been expanded. “Implicit HCI” has been defined as “an action, performed by the user that is not primarily aimed to interact with a computerized system but which such a system understands as input” (Schmidt, 2000). A broad understanding of the term “context” is advised, including the environ ment, situation, state, surroundings, task etc. The author suggests the use of sensors to equip devices with perception capabilities and provides “active environments” with IR networks as an example – an anticipation of the IR-based motion recognition of the CAAS described in this work (see chap. 5.4 Model for Context-Aware Assistive Systems). 7 First published 1991 by Scientific American, Vol. 265, No. 3, pp. 94-104. 34 Background: Computer Science Later the concepts of “embedded interaction” and “implicit use” emerged – they imply that information is embedded into people’s environments (Schmidt, Kranz, & Holleis, 2005). The authors describe the unobtrusive integration of context-specific infor mation on displays in everyday contexts like wardrobes or even umbrella stands. Thus using everyday motions in work environments to implicitly interact with devices and projecting information directly into these work contexts (see chap. 3.2 Projection) are logical advancements of existing HCI lines of research. Although the concept of implicit interaction was influential in research it took some years until the small computers and sensors required reached a broad audience. Like often when information technology crosses the border from specialized applications for industry and research to the mass market, the game industry was a driving force. Nintendo’s Wii released in November 2006 in Japan marked a breakthrough: its wire less primary controller, the Wii Remote, can measure the acceleration along three axes using an accelerometer (combined with an optical sensor). Thus it can be used to in teract with items on screen via gestures and pointing. This form of interaction is more direct and “natural” than using a mouse or a keyboard – for this reason it is called “natural interaction” (NI). Soon these gaming technologies were used and adapted by researchers and therapists for assistive systems (see chap. 3.3 Gamification). Figure 12: Consoles Wii with Wii Remote (left) and X-Box 360 with Kinect (right). Four years after the release of the Wii another controller for a gaming console, this time Microsoft’s XBox 360, repeated this revolution in HCI: Microsoft’s Project Na tal and the launch of the Kinect. In November 2010 the breakthrough was the capa bility to interpret three-dimensional human body movements in real-time. While the Wii still needed the Wii Remote, the Kinect made the human body the controller, so the requirements for implicit interaction reached the mass market. Background: Computer Science 35 Since this technology plays a crucial role in the development of process-oriented and context-aware assistive systems it is described in more detail in the related works sec tion (see chap. 3.1 Motion Recognition). These and other technological advances in the recent years have a paradoxical effect: while they bring the computer closer to the human they also increase the technological distance8. This new situation is a result of the more natural interaction. Steering an application by gestures allows the computer to become invisible or ubiquitous which essentially is the same. When Weiser puts forward the argument that “sociologically, ubiquitous computing may mean the decline of the computer addict” (Weiser, 1999, p. 10) he is right regarding computer usage: easy and intuitive use of today’s devices like smartphones makes knowledge of computers an option instead of a requirement. Accordingly the ideal assistive system – especially for impaired persons – is ubiqui tous and controlled by natural interaction (see chap. 4 Requirements and chap. 5 Model and Architecture). 2.5.2 Assistive Technology Assistive technology is not confined to persons with impairments. Depending on how broadly the term is understood, cars are assistive systems allowing us to move faster than by foot. The use of tools to increase or “augment” (see next sub-chapter) human abilities has always driven technological development. Even in a narrower sense to day’s widely used driver assistance systems like route guidance systems or head-up displays (HUDs) re-define the design space of cars for un-impaired users (Kern & Schmidt, 2009). However, within this work we focus on the research on assistive sys tems for persons with impairments. The established disciplines for research on elderly persons or on persons with impair ments and disabilities (for the relation between elderly and impaired persons see chap. 2.2.3 Aging and Disability) are medicine, psychology or social pedagogy. Assisting these target groups often implies combining technologies: from the wheelchair in vented 1933 to today’s augmented workplaces, interdisciplinary approaches were used sooner and more frequently than in other domains. 8 The lack of a physical representation (which can also be reached by miniaturization) makes the computer technologically less accessible beyond the confined borders of the running application. 36 Background: Computer Science In computer science, research on assistive technologies has been institutionalized in special fields of human-computer interaction (HCI) and related sub-disciplines. The terms used in this field are “designing for accessibility”, “designing for diversity9”, “universal design” or “universal access”. The development of more accessible solutions has been furthered by legal regulations like the Section 508 of the Rehabilitation Act or the 1990 Americans with Disabilities Act (DeLeire, 2000; Lazar & Hochheiser, 2013). However, concepts like universal design are not restricted to impaired persons – in fact requirements associated with impairments like the reduction of the cognitive load have been imperative in HCI without having special user groups in mind (see chap. 4.1 Standards for Designing Interactive Systems). Thus it is not surprising that within both major organizations for research in computer science, the Association of Computing Machinery (ACM) and the Institute of Electrical and Electronics Engineers (IEEE) there are long traditions of research on assistive technologies in various groups and conferences. The ACM features the ACCESS special interest group (Assistive technologies for persons with disabilities) founded in 1973. The ACM conferences ASSETS (Comput ers and Accessibility, since 1994) and PETRA (Pervasive Technologies Related to Assistive Environments, since 2008) both focus on assistive technologies. Although the IEEE is more focused on technology there are several of the currently 38 societies related to assistive technologies, mainly the Robotics and Automation Society (since 1984), the Engineering in Medicine and Biology Society (since 196310), the Systems, Man & Cybernetics Society (since 197211) and the Society on Social Implications of Technology (since 1972). In the last years, the potentials of assistive technology grew exponentially: the success of ubiquitous computing and the accompanying price drop of sensors like accelerom eters or motion detectors allowed new applications. Like the virtual factory (see chap. 2.6.3 Digital and Virtual Factory, Cyber-physical Systems) can potentially assess al most all product parts and machines within a plant by using sensors, the activities of 9 In this context the term “diversity” also includes groups like children. 10 The predecessor was the IRE Professional Group on Medical Electronics which formed in 1952 and merged with AIEE's Committee on Electrical Techniques in Med icine and Biology in 1963. 11 The earliest predecessor was the IRE Professional Group on Human Factors in Elec tronics, which formed in 1958. This name was changed to IEEE Professional Tech nical Group on Human Factors in Electronics in 1963, to IEEE Human Factors in Electronics Group in 1964, to Man-Machine Systems Group in 1968 and to Systems, Man and Cybernetics Group (1970). Background: Computer Science 37 an impaired person can potentially be monitored in a very high level of detail. “Assis tive Cyber-physical Systems” (assistive CPS) have already been proposed (Makedon, Le, Huang, Becker, & Kosmopoulos, 2009) and they aim to integrate heart rate, blood pressure, eye movement, metabolic work demand, mood or affect and compensatory movements. While the CAAS prototype described in this work (see chap. 6 Implementation) mainly focuses on integrating motion depth sensors in the area of production environ ments, the underlying model (see chap. 5 Model and Architecture) describes a holistic sensor-based analysis of work behavior. 2.5.3 Augmented Reality All forms of assistive systems described in the last sub-chapter try to “augment” (from Latin augmento: improve or supplement) the reality of their users. However, aug mented reality (AR) implies that real-world elements are augmented by computergenerated elements. For this reason a wheelchair is not considered AR. The term is linked to both Virtual Reality (VR) and Mixed Reality (MR). In opposition to AR, VR aims to replace the real world completely by a computer-generated repre sentation, preferably in 3D. Mixed reality describes the “virtual continuum” (Nilsson & Johansson, 2008) from augmented reality to “augmented virtuality” as shown in Figure 13: Figure 13: Virtual continuum according to Nilsson & Johansson 2008. Like VR historically AR is focused on graphics. Established technologies like headmounted displays (HMD) or special eyeglasses primarily developed for VR reflect this tradition. One of the first researchers who put forward the idea of an AR “that 38 Background: Computer Science enriches rather than replaces the real world” and “annotates reality” was Steven Feiner (Feiner, Macintyre, & Seligmann, 1993, p. 53). He presented KARMA (Knowledge based Augmented Reality for Maintenance Assistance), a system for maintenance and repair tasks based on a head-mounted, see-through display. This strong and persistent association to graphics and especially to 3D graphics links AR to fields where spatial orientation and perspective play an important role – like architecture, design, navigation, military or medicine. The design of plants also has been an area of interest which links AR to the concept of the digital factory in the sphere of industrial production (see chap. 2.6.3 Digital and Virtual Factory, Cyber physical Systems). The implementation of sensors like accelerometers or global positioning systems (GPS) in handheld devices opened new areas for AR like games, personal navigation or touristic applications. Similarly the rise of motion sensors using depth data (which undoubtedly will soon be integrated in handheld devices) simplified the requirements for interaction with AR because no data gloves or markers are needed to track the user’s movements. Context-aware assistive systems (CAAS) as presented in this work use sensory data (in the current implementation video and depth data, see chap. 6.3) to enhance reality either on a visual level by projecting instructions directly into the workspace (see chap. 6.4.1 Visual Output and Projection) or on a psychological level by adding mo tivating elements with a gamification component (see chap. 6.5). Especially the latter, i.e. the computer-based augmentation of the emotional state of workers, is a new ren dition of AR. The use of projections is much more established and is described in some detail in the related works section (see chap. 3.2). 2.5.4 Artificial Intelligence Artificial Intelligence (AI) is a vast area of research. It originated in parallel to the establishment of the discipline of computer science and is strongly connected to its founders Alan Turing and Marvin Minsky. It remains a philosophical question if a computer will ever be “truly intelligent” in the sense that it can pass the Turing-Test, i.e. completely simulate or mimic human intel ligence. Nevertheless AI has attracted thousands of scientists and as a result special ized systems have long started to be applied in various domains. Most of these systems only use a subset of the traits considered necessary for “true” intelligence: reasoning, knowledge, planning, learning, communication, perception and manipulation of ob jects (Luger, 2009). Still these systems are useful und successful in their domain – be it chess, path finding or enemy strategies in computer games. Background: Computer Science 39 Assistive systems or “wizards” are an example of such a specialized area. If they are used in restricted contexts, the fraction of the real world that has to be modeled for the AI can be narrowed down which makes it more efficient. Navigational systems in cars successfully apply path finding to the real-world by combining map data with position data and traffic data. Although they are very successful many drivers complain about their intrusiveness. This is a result of their limited capacities: if the systems would incorporate learning and optical indoor sensors they could recognize the driver and adjust the level of guidance according to his or her preference and knowledge of the region. This ability to scale according to their user is a challenge for AI. When assistive systems are to become “companions” as proposed in the special re search field “A Companion-Technology for Cognitive Technical Systems”12 the abil ity to scale is essential since companions need to map human cognitive abilities. The formal approach taken for the companion technology was the combination of two planning techniques: partial-order causal link planning (POCL) and hierarchical task network planning (HTN) are joined to “hybrid planning” (Biundo, Bidot, & Schatten berg, 2011). Although these planning techniques are designed for applications work ing with humans, their robust implementation in sensor-based assistive systems remains a future research task. On the other end of formal complexity, the applications most dedicated to user expe rience (UX) encounter the same AI-challenge: games need to scale according to their users. This shows an important parallel between games and assistive systems: they both do not strive for the best solution but for the solution best adapted to the user. Accordingly, implementing a game AI far superior to the current player would result in a feeling of losing and thus counter the desired flow state (see chap. 2.4.3). In par allel an assistive system not trying to detect the user’s capabilities will soon underchallenge or overchallenge his or her skills and thus reduce the acceptance rate. The CAAS framework assumes that the self-regulating or autonomous systems de signed to model players can be adapted for assistive systems (see chap. 5 Model and Architecture) and the comparatively simple13 AI-strategies for goal arbitration in games (Buckland, 2005, pp. 398–414) can be adapted for goals related to service or work processes. 12 German original: “Sonderforschungsbereich sfb transregio 62: Eine CompanionTechnologie für kognitive technische Systeme”. 13 When compared to multi-dimensional formal approaches like hybrid planning. 40 Background: Engineering and Production Management 2.6 Engineering and Production Management In engineering and production management the use of task-specific tools and assistive systems has a long tradition: when working with heavy weights or dangerous sub stances every healthy persons is “impaired” in the sense that the normal capabilities do not suffice and various augmentations are need. Since it has always been in the best interest of production companies to minimize accidents and maximize the work ers’ performance, several concepts and tools used in assistive technology and com puter science originate from the fields of production and automation. 2.6.1 Measuring Work Processes with MTM While Methods-Time Measurement (MTM) clearly draws on the concept of measur ing human behavior established in the social sciences and psychology, the method is situated in the field of engineering and production management. The underlying con cept of measuring human movements in production environments probably goes back over a hundred years to the time Henry Ford14 successfully standardized work pro cesses in the production of Model T by using moving assembly belts (Nevins, 1954). Like its predecessors Motion-Time-Analysis (MTA) or Predetermined Motion Time Systems (PMTS), MTM analyzes and measures work processes. The first extensive system was PMTS or “WORK-FACTOR” (Quick, Duncan, & Malcom, 1962). It was published in 1938 as a response to 1930's prohibition of stop-watches as a means of measuring work. Later MTA and PMTS were replaced by MTM which was first de scribed in 1948 (Maynard, Stegemerten, & Schwab, 1948). The current state is de scribed by REFA (the German Verband für Arbeitsgestaltung, Betriebsorganisation und Unternehmensentwicklung)15 and the USA / Canada MTM Association for Stand ards and Research. MTM breaks human movements down to basic elements like: reach for an object or a location grasp an object , touching it or closing the fingers around it move an object a specified distance to a specified place re-grasp an object in order to locate it in a particular way release an object to relinquish control on it 14 More realistically the concept was developed by Ford’s employees, namely Clar ence Avery, Peter E. Martin, Charles E. Sorensen and C. Harold Wills. 15 In opposition to MTM where the standard times are defined in tables based on mo tion analysis, the REFA approach originally aims to negotiate and adapt the times in a dialogue between employers and unions. 41 Background: Engineering and Production Management To speed up measurement times, MTM can be realized in three levels of detail. These are achieved by combining, statistically averaging, substituting and / or eliminating certain basic motions. An example of simplification in the second level is combining the elements reach, grasp and release to produce the new MTM-2 element get. MTM Increasing level of detail MTM-2 MTM-3 Figure 14: Three systems of MTM based on various levels of detail. The following extract from an MTM-analysis shows the first seven basic elements. Table 3: Example of MTM analysis for basic movements. El. Description LH TMU RH Description 1 Move hand to container R14C 15.6 R14B Moves hand to container 2 Grasp first transformer G4B 9.1 G1A Grasps transformer 3 Move hand clear of container M2B -- -- Holds in box 4 Move transformer to area M10B 16.9 M14C Transformer to plate One Time Measurement Unit (TMU) is one hundred-thousandth of an hour, i.e. 0,036 seconds. The codes in the LH and RH columns refer to those in the MTM time tables, e.g. R14C translates as “reach 14 inches to an object jumbled with other objects in a group, so that search and select occur”. As Deuse explains, all MTM-time values are based on a standard performance iden tical to the REFA performance index of 100% (Deuse, 2010). Since many users of context-aware assistive systems will be physically or mentally impaired, it is not pos sible to use the regular values of the MTM-framework. It has to be adapted to fit the skills of impaired users (see chap. 2.2 Demography and Targeted Users). 42 Background: Engineering and Production Management There has been an attempt to apply MTM to disabled persons called GMTM (from German: “Geschicklichkeits-MTM” or translated: skillfulness-MTM) published in the early eighties (Dieterich, 1983). It refers to Mink’s fundamental work started in the seventies (Mink, 1975) but has not been influential in the domain of work for disabled persons like sheltered work organizations. MTM is also interesting regarding the history of ideas: over fifty years after its first description similar formalized approaches were “re-invented” to describe the interac tions between humans and computer interfaces, e.g. the GOMS model (Goals, Oper ators, Methods, and Selection rules) (Card et al., 1983) or later the simplified keystroke-level model (Kieras, 1993). Although the codes for transcribing actions dif fer, the idea of atomizing human behavior in order to make it describable is identical (see chap. 2.5.1 Human-Computer Interaction). 2.6.2 Assembly Workplaces Manual assembly has always played an important role in industrialization. The MTM approach to analyze human movements in work shows the high granularity which has been achieved in this area. Nevertheless manual work is generally slower and prone to human errors, so assembly workplaces usually are replaced by automated systems as soon as it is economically feasible. However, the growing demand for customized products also increased the need for manual production to the point where we can witness the return of “manufactories” (see chap. 1.1.1 Automation versus Manual Assembly). Thus in a time of digitaliza tion, virtual factories and cyber-physical systems (see chap. 2.6.3 Digital and Virtual Factory, Cyber-physical Systems) the demand for manual assembly work and the cor responding workplaces paradoxically grows. The typical design of an assembly workplace is based on the definition of the activity of assembling. According to the authoritative VDI 2860 on Technology for Assembly and Handling (VDI Verein Deutscher Ingenieure, 1990) the process consists of five activities: assembling or joining (e.g. screwing, plugging, gluing) handling (e.g. grabbing, placing) checking (e.g. measuring, visual inspection) adjusting (e.g. setting, tuning) auxiliary functions (e.g. cleaning, labeling) These activities all involve the human hands, arms and eyes so ergonomics play an important role. Background: Engineering and Production Management 43 Assembly workplaces are usually produced using aluminum profiles. They integrate lamps and power supplies and support work plates made from various materials, de pending on the area of application (see Figure 15). Figure 15: Industrial assembly table constructed by the Schnaithmann GmbH. Especially in manual assembly scenarios the workplaces need to offer access to vari ous “small load carriers” which contain product components like screws, seals etc. In the context of this work these carriers will simply be referred to as “boxes” or “containers”. This is a simplification because for production management and trans portation issues their sizes and other properties actually are defined in great detail by a DIN standard (DIN EN 13199-1:2000-10, 2000). This allows the boxes to be easily integrated in Kanban-chains or automatic feeding systems. 44 Background: Engineering and Production Management 2.6.3 Digital and Virtual Factory, Cyber-physical Systems For a long time mechanical engineering and computer science or software engineering evolved rather isolated from each other. The potentials of computer miniaturization and their implementation in classical engineering products like cars or machines have long been under-estimated (see chap. 3.4 Assistive Systems at Workplaces). A notable exception is the “digital factory”, a concept which came up in the nineties. The VDI (Verein Deutscher Ingenieure, Association of German Engineers) defines it as follows their guideline on “Digital Factory Fundamentals”: The Digital Factory is the superordinate concept for an extensive network of digital models, methods and tools – amongst others simulations and 3 dimensional visualization – which are integrated by a continuous data man agement. The aim is holistic planning, evaluation and the continuous im provement of all essential structures, processes and resources of the real factory in combination with the product.16 VDI-Guideline 4499 (VDI Verein Deutscher Ingenieure, 2008) With simulation and 3D-visualization the digital factory integrates two major areas of computer science. The ideal of a “continuous data management” implies the use of networks and possibly server technologies so a third realm of informatics is integrated. This inter-disciplinary approach made the digital factory an exotic topic for “regular” engineers while at the same time creating a field for co-operation with informatics. The approach dates back to the first computer-based systems for production planning and steering (PPS) which were established in the seventies. At about the same time the steering of machines by numerical control (NC) operations was established which later became CNC (Computerized Numerical Control) – today’s standard in humanmachine interaction (HMI). The combination of PPS and CNC in principal allowed “digital production”. The rise of Computer Aided Design (CAD), Computer Aided Manufacturing (CAM) and Computer Aided Planning (CAP) in the eighties contrib uted the visualization aspect to the digital factory approach. Until today many of the tools developed for the digital factory are not compatible – for that reason various software solutions support different aspects of the digital fac tory, ranging from plant design down to the simulation of NC-operations for a single machine. The two most established software solutions are Tecnomatix by Siemens 16 Author’s translation. German original: Die Digitale Fabrik ist ein umfassendes Netzwerk von digitalen Modellen, Methoden und Werkzeugen – u. a. der Simulation und der 3D-Visualisierung. Sie ist gekennzeichnet durch ein durchgängiges Datenma nagement Ziel ist die ganzheitliche Planung, Evaluierung und laufende Verbesserung aller wesentlichen Strukturen, Prozesse und Ressourcen der realen Fabrik in Verbin dung mit dem Produkt. Background: Engineering and Production Management 45 (incorporating Process Designer, Process Simulate and Plant Simulation) and Delmia by Dassault Systems. Both solutions describe themselves as crucial for product-lifecycle-management (PLM), i.e. the process of managing a product’s lifecycle from the conception through design and manufacturing to service and disposal. In this holistic approach a digital factory is the natural starting point for a product which is intended to be transparent throughout its lifetime. A similar and partly overlapping approach is enterprise re source planning (ERP) which focuses more on the resources involved in a product’s design, manufacturing and sale. To allow any kind of computer-based simulation or visualization, the plant, the ma chines and the product data have to be mapped in detail. This usually is achieved by aggregating existing data and performing additional value stream mapping. Figure 16 shows the relation between different temporal states and digital representations of a factory as well as the relevant processes and tools: Figure 16: Tools and processes for the Digital Factory and the Virtual Factory. Adapted from Westkämper 2006. Additions in red. In the context of this work the most interesting aspect in the diagram is the “Virtual Factory”. It was introduced by Engelbert Westkämper, a renowned expert in the field of production research (Westkämper, 2006). While the digital factory strictly speaking 46 Background: Engineering and Production Management is just a “snapshot” of an existing factory (although the modeled data can be used dynamically for simulations) the virtual factory continuously integrates data from the real factory. In this aspect the virtual factory is a predecessor of what is currently called cyber physical systems (CPS): physical entities which are controlled by computers and in teract with each other17, or as the pioneering Edward Lee describes the concept: “em bedded computers and networks monitor and control the physical processes, usually with feedback loops where physical processes affect computations and vice versa” (Lee, 2008). Although CPS rely more on agent-oriented processes than the original concept of the virtual factory, this feedback loop is exactly what separates the digital factory from the virtual factory. As long as the flow of data is not optimized by cyber-physical systems and context-aware assistive systems as described in this work, the digitaliza tion of production faces two major challenges: Current data management systems already use RFID-chips to determine the location of products or product lots. The real-time combination with ma chine data would potentially allow to assess an individual product’s state at any time – however many machines do not supply that kind of data. If data is provided it often is more machine-specific than product-specific. Also the granularity does not fit the high demands, since often only lots or even com plete orders are tracked. Humans in manufacturing contexts result in high variance regarding both the quantitative and qualitative results. Current approaches to digital factory mostly just handle humans similar to robots. The Process Designer incorporates a generic human model called Jack (and a corre sponding female version). The aims of the resulting “human resource simulation” (Kühn, 2006) are: a detailed design of manual operations, checking the feasibility of tasks, ergonomic analysis, time analysis and generating work instructions. To achieve this, the digital representations of workers have envelopes representing their field of vision and their maximal grasp range: 17 Although CPS are predominantly used in industrial contexts (especially in manu facturing and automotive) the approach can be applied to other areas including healthcare, transportation and even “living”, i.e. consumer products like fridges or washing machines. Background: Engineering and Production Management 47 Figure 17: The software Process Designer within Siemens Tecnomatix allows to create detailed simulations of production processes. Integrated MTM-values can be assigned to the work operations (see chap. 2.6.1 Meas uring Work Processes with MTM). While this allows a process modeling and time estimations based on the pre-defined times, the system lacks adequate sensors to feed information back into the life PLM or ERP systems so that the digital factory “gets populated by living workers” and continuous product processing data and thus be comes a virtual factory. An assistive system that is aware of the state of products in interaction with the worker could help to make production work more transparent and play a significant role on the way towards a virtual factory. The context-aware assistive systems (CAAS) de scribed in this work use sensors that potentially allow the expansion of the model of the digital factory towards a virtual factory so that it includes human workers (see chap. 3.1 Motion Recognition). The integration of “human resources” into the ERP system or related systems today is a technological possibility. While this technological advancement opens new potentials for assistance, quality control and continuous data management, it also raises new ethical questions (see chap. 8.2 Ethical Implications: Towards Humane Work). 48 Related Work: Motion Recognition 3 RELATED WORK 3.1 Motion Recognition Keyboard, mouse and various game controllers have been the ways of interaction be tween humans and computers (HCI) for decades. In using the human body as a con troller, robust interactive body tracking establishes a new form of “natural interaction”. By using the un-augmented human body as a controller, a simple and direct way of interaction is realized that requires little training. The technical development of markerless tracking has been greatly simplified by the introduction of real-time depth cameras for human body tracking (Knoop, Vacek, & Dillmann, 2006; Siddiqui & Medioni, 2010). However, until recently even the best solutions required multi-camera setups, wearable sensors (Heinz, Kunze, Gruber, Bannach, & Lukowicz, 2006) or markers as well as powerful hardware to allow high performance motion recognition. As described by Shotton et al. (Shotton et al., 2011) it was not until the launch of Microsoft Kinect in late 2010 that any solution ran at rates allowing real-time interactions on consumer hardware while being able to handle a full range of human body shapes and sizes undergoing general body motions. Both the Microsoft Kinect (released November 2010) and the similar Asus Xtion (re leased September 2011) draw their capabilities from a chipset developed by the Israeli company PrimeSense18. However, the Kinect combines an infra-red (IR) laser projec tor with 3D Audio and an RGB camera while the Xtion focuses solely on IR. Figure 18: The Microsoft Kinect (bottom) and the Asus Xtion (top). 18 PrimeSense was bought by Apple for $350 million on November 24 in 2013. Related Work: Motion Recognition 49 It is a common misunderstanding that the Kinect uses a time-of-flight camera (ToF), i.e. a camera measuring the duration a light signal needs to travel between the camera and the subject – the light transit time. In fact the Kinect uses a structured light sensor (or as PrimeSense calls it: “light coding”). The technology is based on an IR laser that projects a static pseudorandom pattern of infrared-points (IR-points) on the environ ment. The 3D scene is generated by stereo triangulation which requires two images. The first image is captured by IR sensor, the second is “virtual” because it is based on a hardwired pattern in the IR laser. Since there is a distance between laser and sensor (and both are stable and aligned), the images correspond to different camera positions and allow stereo triangulation (Ten, 2010). Since PrimeSense developed the depth sensor and the IR-technology for the Kinect and the Xtion, their NITE middleware (Natural Interaction Technology for End-user) is used by both sensors. While it is aimed at Java users (including Java Wrappers), Microsoft developed its own middleware called Microsoft SDK aimed at C# develop ers. NITE consists of an application programming interface (API) and a visual pro cessing software described in detail by Ballard (Ballard, 2011). The middleware processes the depth and image data and translates them into mean ingful data, mostly by identifying the skeleton joints. Thus it supports interaction sce narios like user identification and gesture control. The algorithms in NITE are based on the cross-platform, multi-language standards defined by OpenNI, an organization established by PrimeSense to promote a standard API and to ensure compatibility be tween various natural interaction devices. The aim is a single API applicable for all NI devices. A limitation of both sensors is their inability to resolve the human hand and to detect fingers – the hand is interpreted as a single point in 3D space19. Figure 19: Softkinetic DepthSense 311. 19 Within 2012 and 2013 several tools for detecting hands based on depth information have been developed or adapted to integrate the Kinect, the Xtion and similar depth sensors. Good examples are TipTep Skeletonizer and Candescent NUI. These tools were not available in stable versions at the time the prototype was developed – and recent lab studies show that they still have tracking and latency issues, probably re sulting from the low resolution of the depth image. 50 Related Work: Motion Recognition In early 2012 we managed to inspect the Softkinetic DepthSense 311 (released De cember 2011). In opposition to the Kinect and Xtion the DepthSense 311 works with a 3D ToF sensing chipset from Texas Instruments. It includes the middleware iisu (derived from the sentence “the interface is you”) which is based on C++ but includes a C# wrapper. Unlike NITE from PrimeSense, it does not natively support Java. How ever it features not only full body tracking but also integrated finger and hand tracking. Thus it seemed an ideal solution for the aim of detecting intricate finger movements as common in manual assembly. The following table compares the specifications of the three sensors: Table 4: Comparison of motion detection sensors available early 2012: Solution Asus Xtion PRO Microsoft Kinect SoftKinetic DepthSense 311 Operating Types structured light structured light, video, audio time of flight, video Operating Range 80cm - 350cm Xbox: 80cm - 400cm PC: 50cm - 300cm short: 15cm - 100 cm long: 150 cm - 400 cm Resolution Depth IR: 320x240 in 30fps IR: 640x480 in 60fps IR: 640x480 IR: 160x120 Resolution Video RGB: 640x320 RGB: 640x480 RGB: 640x480 As the table shows, the SoftKinetic shares a general limitation of all systems on the market in early 2012: the resolution of the depth image. In fact the DS 311 offers only a quarter of the pixels of the depth images created by the Kinect or the Xtion. The resolutions provided by the three depth cameras available on the market are wellsuited for games or activities using gross motor skills rather than fine motor skills. However, for assembly work fine motor skills are needed. As a result in the imple mentation the skeleton joints or tracking of the worker’s hands were not used – instead we resorted to supervising 3D spheres or areas on the workplace (see chap. 6.3.1 Technical Restrictions and Solution). Related Work: Projection 51 3.2 Projection Since it became a technical possibility in the nineties, the idea that computers “weave themselves into the fabric of everyday life” and become ubiquitous (Weiser, 1999) has been the subject of technical studies. An important part of this concept is that computer graphics break the confinement of screens. Projection is the key technology that enables the augmentation of real world objects by computer-generated images, the creation of an augmented reality. In recent years the miniaturization combined with a rapid decline in prices finally allowed applications like mobile phones with integrated projection. However, most current systems are still using projection as an alternative way of vis ualization rather than a new way of interaction. Interaction requires additional track ing equipment like video cameras or even depth-sensing range cameras that work in combination with the projection. One of the first systems combining projection with interaction was the “DigitalDesk Calculator” (Wellner, 1991). In this prototype of tangible interaction the camera “sees” where the user is pointing, and the system performs adequate actions like pro jecting a calculator or reading parts of documents that are placed on a physical desk. It can be seen as an early realization of what now is called “natural interaction” and was an inspiration to the subsequent approaches including ours. However, at this early technical stage the system was not robust enough to make industrial products feasible. Ten years later the 2001 “Everywhere Displays Projector” (Pinhanez, 2001) was an other approach to make office rooms interactive. The device used a rotating mirror to steer the light from a projector onto different surfaces of an environment and em ployed a video camera to detect hand interaction with the projected image using com puter vision techniques. It was envisioned as a permanent setup for collaborative work in meeting rooms. Even ten years after the “DigitalDesk Calculator” the system still required a high amount of calibration for the projection and especially the interaction component, so again robust industrial application was not economical. Nevertheless the time was ripe for camera projector systems – in 2003 a workshops series started: the IEEE International Workshop on Projector-Camera Systems (Pro cams). This series focuses on systems that combine controllable lighting systems with light-sensing devices. Starting with projection the focus broadened and today includes 3D scanning, flexible display walls and novel display interaction. In 2004 a system more robust than the Everywhere Displays projector allowed direct manipulation of digital objects by combining hand tracking with projected interfaces (Letessier & Bérard, 2004). Although it was confined to a planar display surface, this simplification allowed a latency below 100 ms on regular computers. 52 Related Work: Projection In 2009 the “Bonfire” system (Kane et al., 2009) demonstrated how robust real world augmentation had become within only a few years. By attaching two camera-projec tor-units, the display area of a laptop was extended to both sides by an interactive projection area, allowing users to employ multi-touch gestures and even interact with objects. The projection worked in environments that vary in their physical character istics and lighting. In 2010 a novel algorithm using a depth camera as touch sensor for arbitrary surfaces was presented (Wilson, 2010). It allows to interact with projected content without instrumenting the environment. This algorithm was improved in the UbiDisplays toolkit (Hardy & Alexander, 2012) by clustering points based on neighbor density. These advances are the pre-requisite for surfaces with projections to become as re sponsive to touch as the capacitive displays used in mobile devices. It is used in cur rent research based on this work (see Figure 20 and chap. 8.3.1 Technical Perspectives). Figure 20: Projected assessment game based on the UbiDisplays framework. Lately pico projectors have been used as mobile displays and interaction devices or “light beams”, sojourning everyday objects in a beam and thus turning them into pro jection surfaces and tangible interaction devices (Huber, Steimle, Liao, Liu, & Mühl häuser, 2012). The focus of these developments mostly has been office use, home use (especially entertainment) or mobile computing. With very few exceptions (see chap. 3.4.2 As sistive Systems Using Projection) the use of interactive projections in production en vironments has not been in the center of computer science research so far. Related Work: Gamification 53 3.3 Gamification Video games have always been designed for accessibility – they “operate by a princi ple of performance before competence [so] players can perform before they are com petent, supported by the design of the game” (Gee, 2007, p. 218). This intuitive and playful approach to interaction is now common in many devices and influences con ventional software. Playful design has reached many parts of society and is frequently used to support disabled and impaired or elderly people (Brach & Korn, 2012; Nunes, Silva, & Abrantes, 2010). Originating from education contexts the term “Serious Games” was established. The difference to regular games is that they promote “serious” purposes, i.e. purposes which are directly linked to the real world – e.g. learning a foreign language or traffic signs. If we follow the philosopher Bernard Suits’ sententious definition of gaming, that “playing a game is the voluntary attempt to overcome unnecessary obstacles” (Suits, 2005), serious games are no “real” games because they have a purpose outside of themselves – they have “necessary obstacles”. This systemic flaw of serious games at the same time is the premise for their combination of gaming and real-world prob lems. According to a pioneer in the field of serious games, Marc Prensky, their two major advantages are (Prensky, 2007, p. 147): contextualizing learning materials and thus add engagement adding interaction to the learning process The hypothesis underlying these claims is that increased engagement leads to in creased learning success – or allows learning success in the first place. In Prensky’s view especially “digital natives” (persons who grew up with digital technology as opposed to “digital immigrants”) are in need of engaging learning experiences be cause the regular methods lack the level of immersion required by this target group. While this assumption surely cannot be applied to all younger persons, at least those with reduced attention spans, reduced working memory or reduced processing abili ties will surely profit (see chap. 2.4.2 Cognitive Workloads). This includes persons with impairments (see chap. 2.2.2) and elderly persons (see chapters 2.2.1 and 2.2.3). Serious games have long become an established field for industry and research. An other term frequently used is “applied games” which often is used for games address ing real-world problems outside of the educational domain. Also the term “games with a purpose” has been introduced. Strictly speaking it refers to games “in which people, as a side effect of playing, perform tasks computers are unable [or unfit] to perform” (von Ahn & Dabbish, 2008, p. 58) so the authors propose a form of crowd-sourcing to address problems unfit for artificial intelligence. 54 Related Work: Gamification Unfortunately the attempts to establish clear boundaries between the terms “serious games”, “applied games” and “games with a purpose” were frequently undermined by their use as synonyms. The latest addition in this list of terms for gaming technol ogies transcending the traditional boundaries of their medium is “gamification” – defined as an “umbrella term for the use of video game elements to improve user experience and user engagement in non-game services and applications” (Deterding, Sicart, Nacke, O’Hara, & Dixon, 2011). Especially in the context of health, gamification already has a long tradition. One of the first applied games was “Re-Mission” developed by HopeLab in 2007 – basically a shooter game, where children with cancer could actively fight against virtual tumor cells. It is a common misunderstanding that playing this game directly improved the children’s health which was neither the case nor the intention. Still the visualization of the abstract threat in the game led to a significantly higher reliability in the chil dren’s medicine intake which ultimately was beneficial in the fight against cancer (Kato, Cole, Bradlyn, & Pollock, 2008). Figure 21: Wii with Wii Remote (left), Wii Remote with Wii Motion Plus (right). In 2007 the “games for health” approach reached a new level with the release of Nin tendo’s Wii (see Figure 21). The video game console uses the accelerator-based Wii Remote20 to detect movements in three dimensions. The accessory Wii Balance Board allows measuring the user's center of balance. This and other new input methods were first looked at critically by the game industry, since “new input paradigms create im mense complications when considering user-centered game design because we must now account for differing talent levels of the individual players” (Pagulayan, Keeker, 20 The concept of steering by motion was successful but the accuracy of the Wii Re mote was limited. For this reason in June 2009 Nintendo introduced an expansion device called Wii MotionPlus that allows more accurate motion detection by combin ing a dual-axis tuning fork gyroscope and a single-axis gyroscope. Related Work: Gamification 55 Wixon, Romero, & Fuller, 2012, p. 800). These complications were addressed and when the Wii Fit exercise game was released in 2008 it contained more than 40 activ ities designed to engage players in physical exercises like strength training, aerobics or yoga poses. With almost 23 million copies sold (end of 2013) Wii Fit even became one the best-selling console games in history. Soon scientist and physicians started to exploit the motion analysis capabilities of the console and its input devices for therapeutic use. The effects of this early iteration of video game devices in health contexts were promising: in an analysis of efficacy be tween traditional and video game based balance programs positive evidence has been shown for using the latter (Brumels, Blasius, Cortright, Oumedian, & Solberg, 2008). For economic reasons there were hardly games developed explicitly for therapeutic use in the first years. Only when the natural audience for game-based approaches (kids and young adults) could be broadened by integrating elderly or disabled persons, the first explicitly therapeutic games were developed. One of the examples focusing on this target group is VI-Bowling, a tactile game helping visually impaired users to sig nificantly increase their throwing skills (Morelli, Foley, & Folmer, 2010). A repre sentative example targeting elderly users is SilverPromenade – a game allowing players to go on virtual walks using the Wii Remote and the balance board (Gerling, Schulte, & Masuch, 2011). When the Kinect was launched in 2010 the technological cycle of adopting and adapt ing video game motion technology initiated by the Wii started anew. The Kinect al ready included the exercise game Kinect Adventures! which utilizes human body motion detection in a variety of minigames (as a pre-packaged game it sold 20 million copies, making it the best-selling game for the Xbox 360). In succession an increasing number of researchers and therapists wanted to make use of the new markerless mo tion tracking capabilities to “gamify” medical treatment. Similar to the situation after the release of the Wii there were no applications deliberately developed for rehabili tation purposes, so they started using commercially available motion-based videogames. An overview of their potential and their limitations is given by Putnam and Cheng who advise to use motion games for patients suffering from brain injuries (Put nam & Cheng, 2013). One of the first motion-based video games deliberately built for therapeutic purposes was developed from 2010 to 2012 in the motivotion60+ project. The aim was to create engaging balance exercises that help senior citizens to prevent falls and thus reduce painful and costly hip fractures. The applications resulting from this gamification of exercises are adequately called exergames. The motion-tracking sensor Kinect was used to allow the elderly users natural interaction with a set of mini games (Korn, Brach, Schmidt, Hörz, & Konrad, 2012). 56 Related Work: Gamification Figure 22: A senior citizen using an exergame developed for balance training. Images courtesy of Wohlfahrtswerk für Baden-Württemberg (left) and Korion GmbH (right). Another form of impairment frequently covered by gamification approaches are strokes, which often imply therapies requiring highly repetitive exercises. An example is Break the Bricks which helps stroke patients to recover their psychomotor abilities (Dömők, Szűcs, László, & Sík Lányi, 2012). The concept of applying gamification to other domains and creating more engaging workplaces has already been described (Reeves & Read, 2009). However, the authors focus on general business processes and do not consider natural interaction. To our knowledge the use of gamification for industrial workplaces has not been described outside of the context of the research presented here. Related Work: Assistive Systems at Workplaces 57 3.4 Assistive Systems at Workplaces Assistive systems have been accompanying work for a long time, from expert dia logue systems in customer care centers to step-by-step instructions on monitors in production environments. While new forms of interaction and assistance are readily adopted in many areas, the pervasion of the industrial domain is slow. Today Human Machine Interaction (HMI) still lacks behind the “regular” HCI: even the most ad vanced computer-based assistive systems in production have been attributed to use “suboptimal concepts of information technology” and thus are seen as unapt to ensure “efficient and ergonomic guidance of assembly workers” (Zäh et al., 2007). This situation has hardly changed in the last years: as a recent Fraunhofer study on HMI explains, from the variety of modern interaction techniques only touchscreens found their way to machine interfaces in production environments (Bierkandt, Preiss ner, Hermann, & Hipp, 2011) – and even with typical industry tools much skill and some manual adaptation is required as soon as interaction methods are to be changed. The discrepancies may partly be a result of the traditional distance between mechani cal engineering and information technology, two disciplines which still are vastly sep arated. While CAD (Computer Aided Design) brought computers into manufacturing companies they were considered tools for construction – and not devices for user in teraction. Computers were considered insecure and unfit for the special requirements in production environments where reliability and security are essential. Figure 23: An implementation of the Poka Yoke concept using an impression. The ambitious aim of production companies is to produce “zero errors” or “total qual ity” (see chap. 4.2.2). This goal is reflected in the concept of “Poka-Yoke” where systems are designed to be “fail-safe” or “mistake-proof”. This is can be realized by organizational methods or physical appliances. An impression that can only hold a part if it is assembled correctly (see Figure 23) is a good example of how Poka Yoke is applied in industrial practice. 58 Related Work: Assistive Systems at Workplaces This seeming obsession with quality becomes more plausible if the potential outcomes of errors in apps for mobile devices or regular business software are compared with the effects resulting from human errors or respectively software bugs in production environments. In this domain errors can immediately result in severe injuries of work ers and in combination with just-in-time production also in substantial financial losses. This is probably the main reason why specialized PLC (Programmable Logic Controllers) for a long time have been preferred over generic computers. Although modern PLC can be programmed with developer software running on computers, their main focus has always been regulating and steering – interaction and especially visu alization were implemented later and have long been considered peripheral. As a re sult most manufacturers are very conservative when changing HMI and prefer “safe and slow” over “new and intuitive”. New forms of HCI are implemented more readily if they have become part of an ac cepted standard like ISO 9241 (ISO/TC 159/SC 4, 2006) which is covering the “ergo nomics of human-system interaction”21. Although this and related standards like ISO 14915 are updated regularly they are not designed to describe very recent approaches: while “guidance on tactile and haptic interactions” was added to ISO 9241 as part 920 in 2009 (ISO/TC 159/SC 4, 2009), motion recognition and accordingly natural inter action based on movement has not been covered so far, although this type of HCI is widely used today. Finally in the domain of engineering, innovative solutions are rather patented than published – so innovative HMI solutions can be seen at fairs but are rarely described in journals or conference proceedings. Some of the more advanced forms of interac tion in assistive systems in work contexts are portrayed in the next sub chapters. 3.4.1 Assistive Systems Using Motion Recognition The well-established “pick-by-light”-systems can detect if a worker picked an assem bly component from the right container. However, they cannot examine if the product component was assembled correctly – or even assembled at all. This binary approach (right pick / wrong pick) is well-established in the industrial domain. Most quality gates check if products or specific product parts like weld seams are “okay” or “not okay”. The possibility to obtain more detailed information about what is going on during the work process is comparatively new. 21 Originally it was titled “Ergonomic requirements for office work with visual display terminals” but since 2006 the standard tries to cover more recent forms of interaction. Related Work: Assistive Systems at Workplaces 59 There are first efforts in the industry to reach a continuous analysis of work processes. A good example is a system based on ultrasonic waves by Sarissa GmbH22. The ul trasonic waves are emitted by “trackers” attached to gloves the assembly worker is required to wear and received by sensors mounted above the assembly table. Figure 24: The Quality Assist System based on ultrasonic waves. Image courtesy of Sarissa GmbH. The system compares the worker’s motions with pre-stored motion sequences in realtime. It offers a high accuracy of 5 millimeters (according to the vendor’s website). With the transmitter’s weight of 40 grams and the receiver’s hemispherical detection of eight meters in diameter the system offers good mobility. Still it suffers from some shortcomings – most notably is the parallel to projection: like with head mounted de vices (HMD) a battery-powered physical device (in this case the transmitter) has to be attached to the users. This raises usability and acceptance issues, especially in pro duction environments and even more so with elderly or impaired persons. While the use of systems using image-based motion detection for the surveillance of large areas in production environments has been proposed (Sardis et al., 2010), the use of motion detection to analyze detailed movements of production workers’ has not been described so far. 22 As common in the domain of engineering, technical details about the system have not been published. The descriptions of this and subsequent systems are based on in terviews, brochures and information on websites. 60 Related Work: Assistive Systems at Workplaces 3.4.2 Assistive Systems Using Projection The visualization of instructions or context-specific information directly in the work space is an important aspect of efficient work in production environments. Most sys tems use regular monitors to achieve this. Two more advanced technologies related to the field of augmented reality (see chap. 2.5.3) are head-mounted displays (HMD) and projections. Although HMDs usually compromise a worker’s freedom of movement, their industrial potential has been stud ied extensively (Weaver, Baumann, Starner, Iben, & Lawo, 2010). However, in direct comparison projection is by far the less intrusive system; but although numerous ap plications for office or entertainment use have been described (see chap. 3.2 Projec tion) this technology has rarely been applied to the industrial domain. Interestingly, another light-based assistance has been playing an important role in pro duction assistance for a long time. Most current assistive systems in assembly use “pick-by-light” – a solution where the next box a worker has to pick parts from is marked by a small indicator lamp attached below and the pick is controlled by a light barrier. A reason for the prevalence of this comparatively advanced and intuitive form of HMI might be that light barriers are integrated as sensors, so this form of assistance could easily be realized using the programmable logic controllers (PLC) common in industry. Figure 25: Pick to light display "PickTerm Flexible". Images courtesy of KBS Industrieelektronik GmbH. The most intuitive form of visual assistance – the projection of relevant information directly into the workspace – is still rarely seen in industrial systems. A patent search Related Work: Assistive Systems at Workplaces 61 conducted by the University of Applied Sciences Esslingen showed three relevant pa tents: Computer Aided Works by the iie GmbH & Co. KG, Germany Intelligent Workplace by the Friedrich Martin GmbH & Co. KG, Germany Light Guide System by OPS Solutions, LLC, USA The system Computer Aided Works is a visualization system offering several inter faces that connect to other assistive systems. The Intelligent Workplace already visu alizes information within the work area using a projector. The Light Guide System seems to be most advanced system used in industry and integrates the two approaches. However, the system still relies on external devices for confirmation of processes like touch monitors. The projections themselves are not interactive. Thus even the cur rently most advanced system used in industry fails to combine projection with inter action as demonstrated by Wellner’s “DigitalDesk Calculator” in 1991 (see chap. 3.2 Projection). The concept of in-situ projection of information in production environments is also explored in research. In automotive manufacturing, the quality of spot welding on car bodies frequently needs to be inspected. Currently most spot welding inspections rely on printed drawings. Zhou et al. describe a system that projects visual data onto arbi trary surfaces. It provides just-in-time in-situ information to a user within a physical work-cell (Zhou et al., 2011). Figure 26: Projection of visual data on welding sports. Images courtesy of Jianlong Zhou and Bruce H. Thomas of the Wearable Computer Lab, University of South Australia. Zhou et al., 2011. Like the Light Guide System this approach attempts to solve one problem (instructions usually require the user to change the field of vision) without addressing a related one (confirmation of processes usually requires the user to change the field of vision). In 62 Related Work: Assistive Systems at Workplaces conclusion the advantages of projected interfaces in the industrial domain still remain to be explored. Apart from the research presented here, to our knowledge the only system currently combining projection with interaction in work contexts (although not in a production context) is an assistive system for guiding workers in sterilization supply departments (Rüther, Hermann, Mracek, Kopp, & Steil, 2013). In such departments medical instruments are cleaned, disinfected and sterilized. The system projects instructions directly into the workplace and assists the workflow. Moreover a depth sensor is used to detect the user’s movements and thus allows a projected user interface. Figure 27: An assistive system in the medical domain using projected interfaces. Image courtesy of Stefan Rüther, Research Institute for Cognition and Robotics (CoR-Lab) The huge advantage in this scenario is that a projected interface can never become “unclean” so sterilizing the table is sufficient. The system was evaluated successfully: While the mean time required and the mean number of minor errors remained almost the same, the mean number of major errors was reduced by almost 63% in comparison to the paper-based approach used before. 3.4.3 Assistive Systems Using Gamification Gamification is not a completely new concept but dwells on established approaches like serious games (see chap. 3.3). However, as described at the beginning of this chapter the requirements for new technologies or concepts to be integrated in produc tion environments are high: ideally the new approach is described in an established Related Work: Assistive Systems at Workplaces 63 industry standard. If that is not the case usually “safe and slow” is preferred to “new and intuitive”. Even if new interaction techniques (currently mainly touchscreens) are implemented, the current assistive systems used in production environments are purely functional: they display assembly instructions to decrease the workers’ cognitive load and reduce the sources of errors like the use of wrong tools. Sine the enormous success of attractive mobile devices was noticed also by the man ufacturers of assistive systems for production, the awareness that user experience (UX) is important grows. However, making work more attractive or “increasing fun” so far have not been design goals for tools used in the industrial domain. For this reason to our knowledge, apart from the research presented here, assistive systems in production have not yet been influenced by gamification. Although this work focuses on the domain of production it is helpful to look at other related domains. There have been several efforts to integrate gamification into work processes, especially in the service sector. It has often been described what “ingredi ents” help to gamify work and thus increase engagement (Reeves & Read, 2009). Their work meticulously maps existing game elements like avatars, leaderboards, lev eling and reputation to general business processes. The approach is straightforward: several decades ago frequent flyer programs23 or management approaches like “management by objectives” already implicitly took a step into the world of games. In gaming, missions and goals need to be stated explic itly to make them transparent for the players and measurable for the software. Thus it is not surprising that gamification was very well received in business contexts: in 2011 the research company Gartner predicted that 70 percent of Global 2000 businesses will manage at least one “gamified” application or system by 2014 (Cowie, 2013). Indeed such predictions or the corresponding high market volumes become more rea sonable, if “employers can use Gamification to incentivize employees by establishing clear goals and rewarding those employees that achieve those goals” (Cowie, 2013). In this very broad understanding gamification is primarily a visualization of manage ment by objectives. Castellan et al. illustrated how gamification can be used in work environments like call centers to help agents and supervisors managing their performance (Castellani, 23 The very first frequent-flyer program was created in 1972 by Western Direct Mar keting, for United Airlines. It already awarded members with special bonuses. In 1979 Texas International Airlines created a program that used mileage tracking to give “re wards” to its passengers. 64 Related Work: Assistive Systems at Workplaces Hanrahan, & Colombino, 2013). They also describe a private social network designed by PlayVox for contact centers including a gamified training system. The system has been advertised by quoting the customer GroupOn Latin America as follows: “PlayVox lets us detect and make a quick diagnosis of underperforming agents or those who ignore certain important procedures in serving our customers.” Here the emphasis clearly lies on making life easier for managers: gamification is used as a tool to find and dismiss underperforming employees. With respect to the goals men tioned in the flow approach (see chap. 2.4.3) this might be seen as a perversion of gamification. However, most games are designed to have both winners and losers – so while the application of gamification in this example surely is unethical, it is not unnatural. The fascination for the gamification approach in the service sector seems to be fueled by the increased measurability. Given the comparatively good measurability of work in production both on the process level (see chap. 2.6.1 Measuring Work Processes with MTM) and on the results level, it is probably due to the different mindset of people working in the industrial domain that gamified assistive systems have not been implemented there so far. As the discussion of the approaches in the service sector shows, new issues arise as soon as gamification is implemented in work contexts. Apart from obvious potentials of misuse, there are structural and even philosophical questions. Recently the concern was raised that replacing intrinsic rewards with explicit ones may in the long run re duce work motivation (Finley, 2012). This shows the ethical dimension of assistive systems in work contexts in general and specifically of those using gamification. This dimension is discussed in greater detail in the next sub-chapter as well as in the chap ters 2.3 Ethical Dimensions, 8.2 Ethical Implications: Towards Humane Work and 8.3.2 Ethical Perspectives. Related Work: Ethical Standards for Assistive Technology 65 3.5 Ethical Standards for Assistive Technology To be useful assistive technology needs to “come close” to the user. The extensive use of sensors described in the previous sub-chapters opens possibilities for a detailed physical and potentially also emotional surveillance and guidance – a quantified self. The potential pervasion of our lives by new sensor-based assistive technology – like the context-aware assistive systems (CAAS) presented in this work – requires that technical possibilities are discussed with experts from other disciplines. The early in tegration of an ethical perspective into the design and development process seems to be an adequate approach. The recently proposed MEESTAR-model is a first attempt to formalize this co-oper ation. It describes the “ethical evaluation of socio-technological arrangements” in the domains care and health (Manzeschke, Weber, Rother, & Fangerau, 2013). Figure 28: MEESTAR-model by Manzeschke et al. 2013. Image courtesy of Arne Manzeschke, TTN-Institute. 66 Related Work: Ethical Standards for Assistive Technology The approach separates three layers of analysis24: the societal, the organizational and the individual. Within these layers seven topics are addressed: care, self-determina tion, security, justice, privacy, participation and self-concept. Based on this analysis, the approach differentiates between four ethical verdicts: I. The application is completely uncritical from an ethical point of view. II. The application is ethically sensitive, however the issues can be addressed in practical application. III. The application is ethically highly sensitive; it either has to be permanently monitored or should not be introduced. IV. The application should be rejected from an ethical point of view. Currently already the distribution of verdicts (one neutral, three critical) shows that the model focuses on negative effects and explicitly states that “positive effects are not offset directly with MEESTAR” (Manzeschke et al., 2013, p. 13). This emphasis on potential negative implications also is a reflection of the model’s focus on care and health: in the domain of work positive effects (e.g. on performance and motivation) would structurally be ranked more important – for example in indus trial contexts slight negative effects have always been tolerated if the overall produc tivity was increased. Still the MEESTAR-model is a first approach to standardize the ethical evaluation of assistive systems. Thus it provides important guidelines that can potentially be adapted for CAAS in the production domain. 24 Author’s translations is accordance with Arne Manzeschke. The original German terms are provided in Figure 28. Requirements: Standards for Designing Interactive Systems 67 4 REQUIREMENTS This sub-chapter is based on the following publications: Korn, O., Schmidt, A., & Hörz, T. (2012). Assistive systems in production environ ments: exploring motion recognition and gamification. In PETRA ’12 Proceedings of the 5th International Conference on PErvasive Technologies Related to Assistive En vironments (pp. 9:1–9:5). New York, NY, USA: ACM. doi:10.1145/2413097.2413109 Requirements engineering in our understanding means bringing the technological pos sibilities together with the user requirements and the task requirements. This process is guided by established standards in human-computer interaction (HCI). Since context-aware assistive systems (CAAS) gather sensitive and detailed user data, ethical questions are also taken into account. 4.1 Standards for Designing Interactive Systems On an abstract level CAAS in production environments just represent a specific im plementation of interactive systems. Thus the first step in their design was inspecting generic HCI standards and guidelines and see how they can be applied to CAAS. We will briefly introduce three influential approaches as systems of reference for the requirements. An established standard in HCI are the requirements or “basics of dialogue design” as described in ISO 9241 part 110: Suitability for the task Self-descriptiveness Controllability Conformity with user expectations Error tolerance Suitability for individualization Suitability for learning A second influential HCI standard are Ben Shneiderman’s “Eight Golden Rules of Interface Design” dating back to the 1987 first edition of the textbook Designing the User Interface: Strategies for Effective Human-Computer Interaction (Shneiderman, 2010): 68 Requirements: Standards for Designing Interactive Systems 1. Strive for consistency. 2. Enable frequent users to use shortcuts. 3. Offer informative feedback. 4. Design dialog to yield closure. 5. Offer simple error handling. 6. Permit easy reversal of actions. 7. Support internal locus of control. 8. Reduce short-term memory load. A third pillar of HCI engineering are Jakob Nielsen’s “Ten Usability Heuristics for User Interface Design” from the influential 1993 book Usability Engineering (Niel sen, 1993): Visibility of system status Match between system and the real world User control and freedom Consistency and standards Error prevention Recognition rather than recall Flexibility and efficiency of use Aesthetic and minimalist design Help users recognize, diagnose, and recover from errors Help and documentation These “basics” or “rules” (which are more elaborate in the original texts than the short versions cited here) clearly and efficiently describe how HCI generally has to be de signed. Many of these standards are self-evident, e.g. suitability for the task (ISO 9241 part 110), “strive for consistency” (Golden Rule) and “consistency and standards” (Usability Heuristic). However, standards change slowly; the ones presented here originate from the late eighties and early nineties. When they were established the authors predominantly formulated them for the systems being used in these years. Thus new forms of inter action like natural interaction (NI) might not be addressed and require alterations – and while the standards provide a proven and tested guideline for HCI, both the intended use and the intended users of CAAS give rise to several constraints and ad aptations. These are described in detail in the following sub-chapters. Requirements: Constraints and Adaptations 69 4.2 Constraints and Adaptations There are three major constraints or challenges when designing CAAS for production environments: the focus of interaction, the demand for total quality in production en vironments and the ideal of universal design or universal access. Each challenge re sults in specific requirements. 4.2.1 Focus of Interaction The standards described in the previous sub-chapter implicitly picture users who con sciously interact with some kind of interface. However, in a production environment the main target of a user’s interaction is not an interface of a software but the current work component. While human-machine-interaction (HMI) plays an important role when steering advanced machines by computerized numerical control (CNC, see chap. 2.6.3 Digital and Virtual Factory, Cyber-physical Systems) manual assembly processes are much less digitalized. According to the authoritative guideline VDI 2860 Technology for Assembly and Handling25 (VDI Verein Deutscher Ingenieure, 1990), the central activities are joining, handling, fitting, controlling and auxiliary functions26 like labeling. Interacting with assistive systems is not one of them, alt hough it might be considered an auxiliary function. Although visual elements like instructions or even technical details can be shown on screens or projections close to the working area, the user’s focus will be distracted from manual work if these elements are too prominent. For CAAS to be useful they have to be discrete and “stay in the back” while they are not needed. This results in the primary requirement: R1: implicit interaction Some form of implicit interaction has already been realized in the industrial domain: there are systems with gloves transmitting the hand’s positions by ultra-waves. While these have certain shortcomings like tracking just one point in the glove’s center (see chap. 3.4.1 Assistive Systems Using Motion Recognition) they surely are a sort of motion tracking. Obviously R1 immediately results in a derived more technical re quirement R1’: 25 Author’s translation. German original: Montage- und Handhabungstechnik; Hand habungsfunktionen, Handhabungseinrichtungen; Begriffe, Definitionen, Symbole. 26 Author’s translation. German original: Fügen, Handhaben, Justieren, Kontrollieren, Hilfsfunktionen. 70 Requirements: Constraints and Adaptations R1’: motion recognition Implicit or natural interaction (NI) allows the user’s regular movements to become the predominant input required by CAAS. The technologies required to achieve this have been described in the state-of-the-art section (see chap. 3.1 Motion Recognition). The implementation of motion recognition automatically addresses several require ments: self-descriptiveness (ISO 9241 part 110), “informative feedback” (Golden Rules) and “recognition rather than recall” (Usability Heuristic); these are adapted to production environments with the following second derived requirement: R1’’: detection of the current work state and speed Just by watching the worker’s picks, motion recognition allows the CAAS to identify the current process within a sequence and thus the product state. Also the time be tween picks can be measured and thus the work speed can be deducted for each pro cess. The question if this information is just used internally by the CAAS or if it is communicated to the worker is addressed later in this chapter (R6: protect the user’s personal data). Next to implicit interaction a way to reduce distraction from the center of interaction and thus “support the internal locus of control” (Golden Rule) is moving elements of interest (e.g. instructions or visual feedback) closer to this center. While it would be disadvantageous to integrate a monitor into the work plate, the use of projection has been extensively described for office environments and recently has also been applied to the industrial domain (see chapters 3.2 Projection, 3.4.2 Assistive Systems Using Projection). This results in the following requirement: R2: projection of information 4.2.2 Total Quality A constraint that originates from the production domain is the goal of reaching 100% quality or “total quality”. While this requirement cannot be met in research prototypes it is important to consider that it is an actual requirement in the industry. So when the “basics of dialogue design” require error tolerance that is something alien to production management. Many suppliers are requested to sign agreements where a quality of 99.9% or above is guaranteed. They face severe penalties if these require ments are not met. So if established control instances in production like cameras are to be replaced or supplemented by CAAS, the error controls offered are essential. Errors are also an issue in the HCI standards described, most elaborately they are ad dressed by Nielsen’s request to “help users recognize, diagnose, and recover from errors” (Usability Heuristic). This results in the following requirement: Requirements: Constraints and Adaptations 71 R3: error detection While it is comparatively easy to use the depth data generated by motion recognition to check if a box was accessed (and thus model state-of-the-art systems using light grids) other processes are highly demanding. For analyzing the intricate movements typical for manual production, CAAS require finger detection. As discussed (see chapters 3.1 Motion Recognition, 6.3.1 Technical Restrictions and Solution) the lim ited resolution of the motion sensors available at the time of the prototype implemen tation left this requirement a subject of future research. A potential alternative or complementation to finger tracking discussed for future re search is the implementation of object recognition: the image of the product assem bled correctly can be compared with the actual product (Funk, Korn, & Schmidt, 2014a, 2014b). However this pattern recognition approach is complicated by several everyday challenges in manual assembly like small objects with minimal differences or heavily jointed products. No matter how the challenge of detecting product-related errors will be solved – R3 results in the following secondary requirement for the sub sequent realization of CAAS: R3’: finger recognition / object recognition A future CAAS with finger and / or object recognition would also address another problem: the “easy reversal of actions” (Golden Rule) is difficult to attain in produc tion where components often are joined permanently. A system that detects errors in the making would offer “error prevention” (Usability Heuristic). Thus it would pre vent products from becoming waste that have already progressed very far in the value chain – since assembly often is the last step before packaging. 4.2.3 Universal Design The general HCI standards also have to be adapted to the targeted users of CAAS – in the first step elderly and impaired persons in production. These workers have specific physical and cognitive needs (see chap. 2.2 Demography and Targeted Users). However, in a second step future CAAS might be used by everyone working in pro duction (see chap. 8.3 Future Research). So although there are constraints resulting from the primary target group, the overall aim is to create a universal design and uni versal accessibility for CAAS. Several important design principles have already been mentioned in the first sub-chap ter. However general standards like “strive for consistency” (Golden Rule) are not explicitly transferred into the CAAS requirements unless they have to be adapted. This is the case for “self-descriptiveness” (ISO 9241 part 110) and “offer informative feedback” (Golden Rule). Since the targeted users – both impaired and un-impaired – 72 Requirements: Constraints and Adaptations do not primarily interact with the CAAS, its information density should remain low. This is described graphically by the desktop metaphor (see chap. 2.4.2 Cognitive Workloads) and summed up by the Golden Rule to “reduce short-term memory load”. This results in the following requirement: R4: adaptation to the user’s competence level Since a user’s time required to complete a certain process is known from R1’’ (detec tion of the current work state and speed) the system allows an automatic measurement of work actions similar to methods-time-measurement (see chap. 2.6.1 Measuring Work Processes with MTM) on a macro level27. Thus the CAAS measures how fast a user works and detects changes in work speed. These changes will then lead to some form of adaptation (incorporating the Golden Rule of “suitability for individualiza tion” and the usability heuristic “flexibility and efficiency of use”). For an adequate scaling the observed changes have to be interpreted correctly – a drop in speed might result from boredom or exhaustion or even a sudden cognitive problem. This process is described in the following chapters: 5.3 Flow Curves, 5.4 Model for Context-Aware Assistive Systems, 6.5.1 Designing Flow and 6.5.2 Implementing Flow. Like every system that needs to analyze, model and predict user behavior, CAAS ben efit from additional data sources. The disambiguation of causes for observed behavior is simplified by meeting the following derived requirement: R4’: detection of excitement level and type Knowing whether the user is excited and whether this excitement is positive (e.g. pride, relaxation) or negative (e.g. overexertion, fear) is extremely helpful when the CAAS is adapting to the user. This requirement could be fulfilled by detecting the user’s heart rate and / or by analyzing facial expressions. In the prototype described here this requirement has not been addressed, so it remains a topic of future research. When it comes to adaptation there is an additional requirement resulting from both ethical and economic considerations: R5: integration of motivating elements Work in manual production is repetitive and demanding at the same time: after dozens of identical iterations of the same production sequence a slight change (e.g. another color of a label) must break the routine immediately, because otherwise an error will occur. Therefore it is economically sensible to change the monotonous nature of this type of work by adding change and challenge. The integration of motivating elements from game design (gamification) will increase the users’ self-efficacy by visualizing 27 In combination with finger detection even MTM-3 could be realized implicitly. Requirements: Constraints and Adaptations 73 the current status of work, thus also addressing the “visibility of system status” as well as the “match between system and the real world” (Usability Heuristics). In a second step gamification potentially will enhance work satisfaction which is ethically desira ble – if the positive influences can be proven to be long-term effects (see chap. 8.3.2 Ethical Perspectives). Several of the presented requirements – from motion recognition to emotion detection – analyze the user in an unprecedented level of detail. While this raises ethical ques tions, the corresponding technical requirement primarily is “simply” data security: R6: protect the user’s personal data While this requirement seems obvious, data security is an issue in a world that strives for “continuous data management” (see chap. 2.6.3 Digital and Virtual Factory, Cyber-physical Systems). As the requirements study shows, most enterprises presup pose that a CAAS will be integrated in the existing data network structures. At the same time they are aware of the fact that process-integrated sensor-based systems pro duce a different quality of data than the current solutions. The requirements are summarized in the table below: Table 5: Requirements for an ideal CAAS. No. Requirement R1 implicit interaction - R1’ - automatic detection of movements - R1’’ - detection of the current work state and speed R2 projection of information R3 error detection - R3’ - finger recognition / object recognition R4 adaptation to the user’s competence level - R4’ detection of excitement level and type R5 integration of motivating elements R6 protect the user’s personal data 74 Requirements: Requirements Study 4.3 Requirements Study The requirements for CAAS could be derived by extrapolating technical trends and adapting existing HCI approaches and solutions to the domain of industrial produc tion. Since many of these approaches were new and alien to this domain, the views and problem awareness of potential stakeholders from the industry were evaluated in a study. 134 industrial companies in Germany have been asked about present and fu ture demands for assistive systems in the industrial practice. 29 organizations (21.6%) answered the questionnaire28. Most companies (in total 18) were small and medium sized with under 1.000 produc tion workers (62%), six were large with 1.000 to 10.000 workers (21%), four were very small with under 100 workers (14%) and one organization was very large with more than 10.000 workers. All companies face the challenge of integrating workers with impairments which are either related to old age or to accidents. Four companies stated that the percentage of impaired workers is above 10%. While this study is too small to be truly representative of the industry as a whole, the fact that already 13.8% of the companies employ more than 10% of impaired persons is highly relevant for the future demand for CAAS in work contexts. The high percentage of impaired workers found in our study is sup ported by the World Health Survey (World Health Organization, 2004) as portrayed in chap. 2.2.2: the average prevalence rate of significant functioning difficulties in the adult population was 15.6% (ranging from 11.8% in higher income countries to 18.0% in lower income countries) and the average prevalence rate for adults with very sig nificant difficulties was estimated 2.2%. So even in a high income country like Germany about 13% of the working population are impaired. Considering the above average risks and above average physical exer tion in production work the percentage of impaired workers will very probably be significant higher in this domain. Also all companies clearly see that the number of elderly and impaired workers will increase further, so the level of problem awareness in the industry is very high. 28 This high response quota is a result of telephone interviews and personal interviews being used in addition to an online survey tool. 75 Requirements: Requirements Study Number of Companies With regard to the reduced working memory of impaired persons (see chap. 2.4.2 Cognitive Workloads) and R4 (adaptation to the user’s competence level) the follow ing question was asked: How many processes or sub-tasks typically have to be com pleted by one worker at a single workplace? 10 9 8 7 6 5 4 3 2 1 0 31%; 9 28%; 8 17%; 5 14%; 4 10%; 3 1‐5 6‐10 11‐15 16‐20 > 20 Number of Processes Figure 29: Number of processes at a single workplace. As Figure 29 shows the number of processes varies strongly. Still in the majority of 31% of the companies the workers have to handle more than 20 processes within a production sequence. However, there are product variations and usually also com pletely different products: in 29% of the companies changes in the work sequence occur hourly, in 24% within a shift, in 14% daily and in 24% weekly. This means that 90% of the production workers will have at least weekly changes, so to work efficiently a worker will usually need a cognitive mapping of several dozen steps. With the goal of producing “total quality” it is not surprising that impaired workers will need assistive systems support to do this demanding work. Also CAAS would probably allow more elderly workers to continue this line of work when mental capacities start to decrease. In spite of their potential benefits, are only 50% of the companies asked are familiar with assistive systems supporting production work. 11 out of the 14 companies aware of such systems also use them – i.e. almost 80%. However, the assistive system used are state-of-the-art systems with little process-orientation and sub-optimal forms of human-computer interaction (see chap. 3.4 Assistive Systems at Workplaces). The companies asked only used assistive systems controlling the workers’ picks from boxes (pick-by-light). More advanced systems like hand trackers based on ultra-waves or systems which use the projection of information were not mentioned. 76 Requirements: Requirements Study Figure 30 shows the features most requested from a new kind of assistive system: 25 Number of Companies 75%; 21 68%; 19 20 57%; 16 50%; 14 15 32%; 9 10 5 0 Control picks. Only Control Detect Train new number of errors. workers. instruct on error. elements picked. Requested Features Figure 30: Requested features of a new assistive system. The most important industry requirement is the detection of errors (R3: error detec tion). However, this requirement is met by current assistive systems only on the level of picks (i.e. a light barrier checks if a container was accessed). For error detection on the product level, motion tracking systems with finger tracking (R3’ finger recogni tion / object recognition) are required. Although such hand tracking systems have been realized using ultra-wave-based tracking (see chap. 3.4.1 Assistive Systems Using Motion Recognition) these solutions do not allow the tracking of fingers or objects. Thus from a technical standpoint the introduction of motion tracking cameras is inev itable. This also concerns the economic sphere: ultra-wave-based systems currently cost about ten times the price of systems with depth cameras. This is especially im portant when considering the finding that 16 of the 25 companies asked (64%) would only be willing to pay between € 3.000 and € 7.000 for a new assistive system (and five companies, i.e. 20% would even pay less). In consequence motion recognition is the only option process-oriented assistance can be realized economically today. Other potential benefits of using motion recognition were highly appreciated: 9 out of 29 companies (31%) found an assistive system that can give advice on ergonomic issues “highly attractive” and 17 companies (59%) considered it “attractive”. When asked about an assistive system that uses motivating elements (R5) like achievements or high scores adapted from game design, surprisingly 3 out of 28 (11%) found this perspective “highly attractive” and 16 (57%) found it “attractive”. So a majority of the production companies was interested both in ergonomic feedback and gamification. Requirements: Requirements Study 77 All these features – even the essential requirement to detect errors – require motion recognition (R1’). However, when asked about “cameras” being installed at work places only three out of 28 (11%) companies found that unproblematic, whereas 11 (39%) found it “critical” and 10 (36%) perceived cameras as “very critical”; four com panies even felt their installation at the workplace was “impossible”. In this context it is important to note that 20 out of 29 companies (69%) found that a new assistive system needs to be connected to the company’s existing data networks like the pro duction planning system, whereas only nine companies (31%) would be fine with a stand-alone system. If the connection to the company network is seen as essential, a camera naturally be comes a great hindrance because employees could unknowingly be visually super vised. Thus for the establishment of CAAS in production a clear technical separation of the user-centered camera-based system from the company’s existing product-ori ented systems will be essential (R6: protect the user’s personal data). Only a system with a “black-boxed” camera will be acceptable and corresponding to ethical guide lines (see chap. 8.2 Ethical Implications: Towards Humane Work). 78 Model and Architecture: Adapted HAAT-model 5 MODEL AND ARCHITECTURE This chapter is based on the following publication: Korn, O., Funk, M., Abele, S., Hörz, T. & Schmidt, A. (2014) Context-aware Assis tive Systems at the Workplace. Analyzing the Effects of Projection and Gamification. In PETRA ’14 Proceedings of the 7th International Conference on PErvasive Tech nologies Related to Assistive Environments. New York, NY, USA: ACM. In the following we describe the model on an assistive system that addresses the re quirements described (see chap. 4): a context-aware assistive system (CAAS) for workers in production environments. The model draws from three existing concepts from different domains: the established HAAT-model for designing assistive systems (see chap. 2.4.1), the concept of flow (see chap. 2.4.3) and the framework for an adap tive game system. 5.1 Adapted HAAT-model The first basis of the CAAS-model is the established HAAT-model (see chap. 2.4.1 HAAT-model) which describes the basic components and functions of an assistive system. The new model focuses on improving three aspects: the human-technology interface is realized by projection the environmental interface is realized by natural interaction (NI) based on human body tracking the activity output is enriched by gamification NI implies real-time motion recognition (requirement R1’) and thus almost real-time intervention – an important requirement in work contexts, where errors can lead to accidents. Using NI also led to an addition to the model (Figure 31: Adapted HAATmodel, green dashed arrow): Since the user’s body becomes the direct controller with out the need for purposeful or even conscious interaction, the system’s environmental interface directly interacts with the user. This implicit interaction (R1) is helpful be cause errors (R3) will often be performed unconsciously and without notice. Model and Architecture: Adapted HAAT-model 79 Figure 31: Adapted HAAT-model. At the same time motion-based implicit interaction opens a channel for continuous performance analysis. Movement is one of the data sources which can be used to model the user’s state and scale the challenge level to meet his or her current level of competence (R4). This is important because especially with impaired users the per formance level can change significantly several times within a single day due to the strong variation of physical and mental states. The scaling of the challenge level is also crucial for the implementation of contextspecific feedback like the gamification component: if activities are designed to create and preserve a flow state (see chap. 2.4.3) they have to adapt to changing competence levels. The idea of adapting the level of challenge to the user is well-established in a domain where flow is not an optional feature but the major aim of the development process: games engineering. An additional advantage is that gaming approaches have always been trying to incorporate and measure the emotional side of the user (Sykes & Brown, 2003) so they directly address the requirement R4’(detection of excitement level and type). 80 Model and Architecture: Framework for an Adaptive Game System 5.2 Framework for an Adaptive Game System The second base of the CAAS-model is an approach to player-centered game design. It aims “to provide a more appropriate level of challenge, smooth the learning curve, and enhance the gameplay experience for individual players regardless of gender, age and experience” (Charles et al., 2005, p. 1). The authors also imply that such adapta tions “decrease task-based failure and error rates among users” which is exactly what assistive systems at workplaces are to achieve. They propose a “potential framework for an adaptive game system”: Models of player types Monitor player performance Player preferences Adapt the game to individuals Measure the effectiveness of adaptation Re-model player types On-line Adaptive game system Figure 32: Potential framework for an adaptive game system by Charles et al. The framework first integrates the player types – typically these are modeled as “Bartle Quotients”, referring to combinations of the four basic types of players (killer, socializer, achiever and explorer) as proposed by Richard Bartle (Bartle, 1996). Player preferences include preferred play styles, e.g. “casual” versus “hardcore” – a prefer ence which can easily be met by adjusting the difficulty level. The player’s performance, i.e. his or her behavior in the game world is then used to adapt the game. This can be achieved by spawning new enemies to increase challenge or by spawning a small group instead of a large one to reduce challenge. As the game allows to assess the results while the player is responding to the adapted challenge it is possible to measure the effectiveness of the adaptation and potentially come to a new model of the player. As an example a player who usually follows an aggressive play style (killer) might suddenly start to gather items (explorer), so the game should adapt to the new play style and spawn more collectables. Model and Architecture: Framework for an Adaptive Game System 81 Clearly this model cannot directly address the situation of CAAS in production envi ronments. Like in the classic HAAT-model the monitoring of the performance (in Cook & Hussey’s terms the environmental interface) in this framework is principally expected to be realized explicitly by mouse and keyboard (or joystick or gamepad). However, this difference is not relevant because games are designed to interact con stantly with the user29 – and since motion tracking ideally provides 30 or more frames per seconds of “performance” in the form of movements, there is no fundamental dif ference in the frequency of interaction. If a system receives real-time or almost realtime feedback, the process of adapting or scaling the output can be almost instantane ous as well. In this respect NI can be seen as an instrument that allows game-like interaction in non-gaming contexts. Especially in manual production, where the main target of interaction is not software but the tools and the product, implicit interaction is the pre-requisite for high-fre quency interaction30. An important aspect of this model is the way it describes the adaptation to the player: an iterative process that leads to a permanent re-assessment of what the user might want and what he or she currently is capable of. This high degree of flexibility and responsiveness is what makes the “Framework for an Adaptive Game System” a suit able reference point for CAAS. 29 Even turn-based games adapt their graphic output to every mouse movement of the player and might even pre-calculate AI responses on the player’s current actions. 30 However, the motion recognition systems available at the time of the implementa tion did not provide the spatial and temporal resolution necessary to interpret intricate finger movements as common in manual production, so the model’s implementation had to be simplified. 82 Model and Architecture: Flow Curves 5.3 Flow Curves A high frequency of interaction combined with an iterative user re-assessment, potentially allows to create a “perfect fit” for the user’s competence level. However, in achieving this, a major aspect would be neglected: to achieve flow, the user has to reach an area between “arousal” and “control” (see chapter 2.4.3 Flow) – but this area is neither a point nor is it stable. To be permanently motivating an activity has to be designed in phases that partly arouse the user and partly give him or her the feeling of control, so that flow comes in waves or curves. The same applies to the gamification of assistive processes: to maintain flow the challenge level has to hover above and below the perfect fit which would be the user’s current performance level. The modeling of the user type in production environments is simplified because there is only one activity related to assembly: building or in Bartle’s terminology “achiev ing”. Thus the player type or the “worker type” can be characterized to a high amount by the frequency and amplitude of the flow curve hovering between the two poles of arousal and control. Figure 33 illustrates this concept: B A Figure 33: Flow curves The two exemplary curves represent two users with different characteristics: User A (blue) needs frequent longer phases of lower challenge level (control or even relaxa tion) whereas User B (red) needs frequent arousal to maintain flow. A major challenge of the flow-oriented approach is interpreting the user’s actions cor rectly: a decrease in performance can be both, the result of boredom or the result of resignation due to overextension. Thus an ideal implementation of CAAS would not only support finger detection but also detect emotions. 83 Model and Architecture: Model for Context-Aware Assistive Systems 5.4 Model for Context-Aware Assistive Systems The CAAS-model combines the advanced HAAT model with the framework for an adaptive game system and the concept of flow curves explained above. It is realized in two levels of detail: a more abstract version (Figure 34: CAAS-Model – abstract) where all sensory inputs and outputs are combined and a detailed version used to ex plain the implementation details (Figure 35: CAAS-model – detailed). Human Cognitive Generator (determines Activity) Cognitive Interpreter Environmental Interface Input Output Context‐aware Assistive System Environmental Interface Output Input Motion Recognition Emotion Recognition Interpreter Audio Haptic Instructions Feedback Generator Model human state physi‐ cal Visual emo‐ tional men‐ tal determine position on flow curve adjust opera‐ tion mode Game Generate adapted interventions Figure 34: CAAS-Model – abstract. On the highest level the model separates the human (green area) and the context-aware assistive system (CAAS, blue). The model aims to show the parallels in processing information. Both the human and the CAAS share an environmental interface consist ing of sensors on the input side and various actors on the output side. The overall aim is that the CAAS input side receives enough data for the interpreter to create a fitting model of the current state of the user. While the physical input can be measured with motion technology, the direct deriva tion of the emotional state from this data only works in obvious cases like trembling or a stiff pose. However, changes in speed within similar tasks can be used to detect a tendency towards arousal or control and boredom. Additional data sources increase the model’s accuracy, e.g. the heart rate or the facial expressions (which both can be extracted from a high resolution video). Hands Figure 35: CAAS-model – detailed. Fingers physi‐ cal emo‐ tional Model human state Interpreter Body Motion Recognition Input Environmental Interface men‐ tal Context‐aware Assistive Sys. Body Physical Output Environmental Interface Heart Face determine position on flow curve Heart Emotion Recognition Face Emotional Cognitive Generator (determines Activity) Human adjust opera‐ tion mode Skin Skin Instructions Monitor Generate adapted interventions Game Generator Projection Visual Output Eyes Physical Input Cognitive Interpreter Feedback Speakers Audio Ears Wearable Haptic Skin 84 Model and Architecture: Model for Context-Aware Assistive Systems Model and Architecture: Model for Context-Aware Assistive Systems 85 The model’s structural analogies continue on the processing side. The human and the CAAS share an interpreter and a generator. The CAAS interpreter uses the data from the environmental interface to model the human state. As an example a detected se quence of errors results in a lower mental score, a detected stress symptom shifts the flow state towards arousal. This model of the user’s current state is then used to de termine his or her position on the flow curve, i.e. to analyze if the current trend moves towards arousal or towards control. This analysis eventually results in an adjustment of the operational mode. This could affect the speed of production, the number of steps to be assembled or even the prod uct. Since a typical phase of a flow curve lasts several minutes, determining the suit able turning point is of essence, e.g. reducing challenge and thus shifting towards control too early will reduce the momentum. If for example the interpreter needs to determine if a worker reduces work speed be cause of boredom or because of exhaustion, specific data reflecting the emotional state (e. g. nervous hand movements, sweat or a fixed gaze) increase the accuracy of the modeled stress level. The behavior after an adaptation of the operation mode will in dicate if the human state was modeled correctly – in the above example increased speed would indicate that the state was interpreted correctly as under-challenge while reactions showing stress symptoms would indicate that the state was misinterpreted and the person was in fact already above the upper challenge limit and outside of the flow channel. Thus the iterative interpretation of behavior changes as results of the adaptions can be used to correct errors in user modeling. The CAAS generator adapts the interventions: the gamification component (e.g. the speed of visual elements or their size and positioning), the instructions (e.g. by in creasing the level of detail in situations of stress or after multiple error occurrences) and the feedback (tone, length and modality). The adapted interventions are then dis tributed over various output channels: monitor or projection and also speakers if au ditory feedback is needed. Potentially the feedback and the gamification can even influence wearables: e.g. a smart watch or a wristband could vibrate when errors are detected. The implementation of the environmental interface and the interpreter is described in chapters 6.3 Motion Recognition and 6.4 Instruction and Performance Analysis, the implementation of gamification is described in chap. 6.5 Gamification Component. 86 Implementation: Model for Context-Aware Assistive Systems 6 IMPLEMENTATION This chapter is based on the following publications: Korn, O., Schmidt, A., Hörz, T., & Kaupp, D. (2012). Assistive system experiment designer ASED: A Toolkit for the Quantitative Evaluation of Enhanced Assistive Sys tems for Impaired Persons in Production. In ASSETS ’12 Proceedings of the 14th in ternational ACM SIGACCESS conference on Computers and Accessibility (pp. 259– 260). Presented at the ASSETS ’12, Boulder, Colorado, USA: ACM Press. doi:10.1145/2384916.2384982 Korn, O. (2012). Industrial playgrounds: how gamification helps to enrich work for elderly or impaired persons in production. In EICS ’12 Proceedings of the 4th ACM SIGCHI symposium on Engineering Interactive Computing Systems (pp. 313–316). New York, NY, USA: ACM. doi:10.1145/2305484.2305539 Korn, O., Brach, M., Schmidt, A., Hörz, T., & Konrad, R. (2012). Context-Sensitive User-Centered Scalability: An Introduction Focusing on Exergames and Assistive Systems in Work Contexts. In S. Göbel, W. Müller, B. Urban, & J. Wiemeyer (eds.), E-Learning and Games for Training, Education, Health and Sports (Vol. 7516, pp. 164–176). Berlin, Heidelberg: Springer Berlin Heidelberg. Retrieved from http://www.springerlink.com/index/10.1007/978-3-642-33466-5_19 Korn, O., Schmidt, A., & Hörz, T. (2012). Assistive systems in production environ ments: exploring motion recognition and gamification. In PETRA ’12 Proceedings of the 5th International Conference on PErvasive Technologies Related to Assistive En vironments (pp. 9:1–9:5). New York, NY, USA: ACM. doi:10.1145/2413097.2413109 The requirements for an ideal realization of a context-aware assistive system (CAAS) have been described (see chap. 4). A model of an ideal system meeting all the require ments has been provided in the previous chapter. Due to the pioneering character of CAAS (at the time of the implementation in 2011 to our knowledge assistive systems based on motion detection have never before been used in assembly contexts) it was evident that not all of the requirements could be addressed in the first implementation: finger detection, object detection (and thus realtime error detection) and the detection of mood31 or excitement were out of scope due 31 While the Kinect camera could be used to apply the Facial Action Coding System (FACS), the required top mounted position did not allow to capture facial expressions. 87 Implementation: System Overview to the limitations of the available sensors (see chapters 3.1 Motion Recognition, 6.3.1 Technical Restrictions and Solution). The implementation of the system and the tech nical challenges faced during that process are described in the following sub-chapters. 6.1 System Overview As Figure 36 illustrates, the CAAS prototype is based on physical components (the experimental assembly table, the motion detection system and the projection system) as well as software components handling the assistance, the gamification and the re cording of user data. These components are described in the following sub-chapters. Physical Components Environmental Interface Experi‐ mental Assembly table Projector and Monitor Motion Recognitio nSystem Software Components Interpeter / Generator Assistive System Experiment Designer (ASED) Designer Gamification Component Runner Figure 36: System overview. 6.2 Physical Integration in the Work Environment The design of the experimental assembly workplace is based on the design of regular assembly tables as currently used in the industry (see chap. 2.6.2 Assembly Work places). In order to test the effects of the two augmentations in an empirical study, the work environment’s complexity was deliberately lowered. The aims were reducing distrac tions and allowing easy transportability while still being able to adjust the work sur face in height to provide equal conditions for wheelchair-bound persons. The resulting experimental assembly table is described in Figure 37: 88 Implementation: Physical Integration in the Work Environment top cage holding the mo tion recognition sensor with video recording and the projector adjustable monitor base able to support a touch screen to map state-of-the art assistive systems work area with boxes for assembly parts Figure 37: Experimental assembly table with motion recognition and projection. Like its industrial counterparts, the system has been constructed to meet the require ments of the authoritative VDI 2860 on Technology for Assembly and Handling (VDI Verein Deutscher Ingenieure, 1990) described in chapter 2.6.2. It was made from alu minum profiles and weighs about 70 kilograms. The base to the left can support a monitor as required to map state-of-the-art assistive systems. The boxes are arranged horizontally to avoid occlusion by other boxes mounted above. The table also had to be able to support a projector for the in-situ projection of instructions and a motion detection system for the recording of the user’s movements. Both are installed in the top cage mounted 1.4 meters above the table’s plate to minimize distraction and provide sufficient lens coverage for the working area. Implementation: Motion Recognition 89 6.3 Motion Recognition 6.3.1 Technical Restrictions and Solution To track the assembly processes, motion sensors are used – and since spatial resolu tion is of essence in tracking, the sensors needed to provide depth information. To decide which sensor was best suited for the prototype we tested the Asus Xtion (re leased September 2011), the Microsoft Kinect (released in November 2010) and the Softkinetic DepthSense 311 (released in December 2011). While the first two sensors draw their capabilities from a chipset developed by the Israeli company PrimeSense and use structured light, the DepthSense 311 uses timeof-flight technology. We decided to use the Kinect because it combined a high reso lution of the depth image (identical to the Asus Xtion and 16 times the resolution of the DepthSense 311) with a video camera (see chap. 3.1 Motion Recognition). As described in the model, an ideal context-aware assistive system would recognize intricate movements including finger movements, it would detect objects and recog nize human emotions. At the time of the implementation (mid 2012) none of the avail able low-priced motion recognition systems did provide the spatial and temporal resolution necessary for the robust implementation of these features. Thus the proto type had to be simplified. The most striking consequence was that no real-time error detection could be realized. A further technical restriction was the impossibility to use the skeletal joint detection as provided by the existing middleware NITE and Microsoft SDK (see chap. 3.1 Mo tion Recognition). Both APIs did not allow to rotate the Kinect or the Xtion because they assume that the sensor will scan horizontal rather than vertical areas. This re striction also applies to vertical mount, i.e. sensors facing downwards on the work space. However for assisting assembly processes a horizontal orientation is impossible32 because the sensor’s opening angle could not scan all relevant areas (de positing racks, working area and the worker’s hands). This led to an architecture where the depth images generated by the sensor software were used as the input while the skeletal joints extracted by the middleware were not utilized. Instead of using the joints we created a new system of reference by analyzing the changes of z-values. We called this the “adjoined spheres approach”. 32 Even the combination of several sensors in a joined 3D space would not have solved the problem, because the worker’s picks – a basic requirement of assistive systems in production solved by light grids in the state-of-the-art – could only be monitored ro bustly from a top view. 90 Implementation: Motion Recognition 6.3.2 The Adjoined Spheres Approach In the adjoined spheres approach movements are defined as passages through adjoined spheres (see Figure 38). By adjusting the spheres’ radius the movement corridors or trigger areas can be designed to fit both the production scenario and the competence of the worker. For example a highly competent worker can be granted wider move ment corridors to avoid early interventions which would otherwise reduce the sys tem’s acceptance. In the following we briefly describe the technical implementation of the adjoined spheres approach33. Figure 38: Illustration of Adjoined Spheres Approach. Although the approach was originally developed using spheres, the algorithm can be adapted for other target areas or geometric bodies like cuboids or cylinders. These target areas are not necessarily empty when the system is initialized – they typically contain parts of objects the worker has to interact with, e.g. the boxes. Therefore it is not sufficient to detect points inside the defined areas – the CAAS has to detect changes with respect to the initial condition inside the area. This can be implemented efficiently by calculating a representative number for the depth values or z-values inside of each sphere – the sum of all contained depth values. For a sphere with the radius ݎaround the central point ܲ௧௧ the calculation is: 33 Patent pending: Method for guidance and/or control of assembly and commission ing processes at workplaces [Verfahren zur Anleitung und/oder Kontrolle von an ei nem Arbeitsplatz auszuführenden Montage- und Kommissionierungsprozessen]. 91 Implementation: Motion Recognition Equation 1: Z-values in sphere. ܼௌ ൌ ∑ ܼ௦௨ௗ | ଶ ଶ ଶ ට൫ݔ௦௨ௗ െ ݔ௧௧ ൯ ൫ݕ௦௨ௗ െ ݕ௧௧ ൯ ൫ݖ௦௨ௗ െ ݖ௧௧ ൯ ݎ Using the reference value ܼோ of the each target area, the sums are permanently recalculated and compared to the reference values: Equation 2: Comparison with reference values. ܼௌ െ ܼோ ܼோ ൌ ܼோ Each target area can have one of three states: occluded, neutral and occupied. The system infers a collision when ܼோ > 0.05. This value proved stable enough in experiments to filter out measurement inaccuracies in the depth images. In an industrial application it will have to be adapted to the products and lighting conditions. A value of ܼோ < -0.05 means that depth values inside the target area are missing – in this situation the system infers that the sphere is occluded. Figure 39 shows the tool developed for testing and optimizing the adjoined spheres approach: Figure 39: Target area states: Occluded (red), neutral (blue), occupied (green). 92 Implementation: Motion Recognition Since the interactive areas work independent of the skeletal joints used by the tradi tional middleware (see chap. 3.1 Motion Recognition) they can also be used for a basic object analysis. Figure 40 illustrates how the depth image allows to infer the threedimensional orientation of an object: Depth image top view Figure 40: Detecting three-dimensional object orientation with interactive areas. A restriction of the approach resulting from the limited resolution of sensors available at the time of the implementation is that small-scale movements cannot be detected, e.g. the number of revolutions when a screw is tightened manually. This would require either a much more refined object interpretation or real-time interpretation of finger movements. These advances – as well as a more detailed object recognition – are sub jects of future research (see chap. 8.3). Implementation: Instruction and Performance Analysis 93 6.4 Instruction and Performance Analysis Our basic research on motion recognition using depth data in combination with the adjoined spheres approach (see chap. 6.3.2) allowed us to implement a system that analyzes passages through interactive areas in real-time. Thus work processes can be analyzed quickly by the interpreter – a prerequisite for context-awareness resulting in adequate feedback or visualization. Since the prototype was primarily used for testing the effects of the augmentations, the software for performance analysis and instructions was called “Assistive Systems Experiment Designer” (ASED). Although it has been developed with production en vironments in mind, ASED can be applied to measure human activities in a wide range of work processes and potentially also in more generic scenarios where coordinated movement is required. ASED implements a designer mode and a runner mode. While the first is used to design the experiments (mapping of work steps, positioning of interface elements) the second documents a running experiment on video and by generating time stamps. Both modes are discussed in greater detail in the following sub-chapters. With respect to the model (see chap. 5) ASED implements the following elements: Environmental interface: input: motion recognition Environmental interface: output: in-situ projection and monitor Interpreter: model human state: physical Generator: instructions The gamification component described later in this chapter handles three additional elements: Interpreter: model human state: emotional Generator: game Generator: feedback Figure 41 illustrates the architecture of ASED which was implemented in Java. Addi tionally the open libraries OpenCV (Open Source Computer Vision Library) and the OpenNI (Open Natural Interaction) have been used. 94 Implementation: Instruction and Performance Analysis Figure 41: Architecture of the ASED software. 6.4.1 Visual Output and Projection Some basic technical functions underlying the system handle the visual outputs in cluding the projection. To avoid discrepancies in the performance of users related to minor interface issues, two elements were projected in all three test scenarios: the controls for “next” and “previous” instruction and the indicators for picks (numbers in front of the boxes emulating the established pick-by-light systems). To display these interface elements the projector had to be active in all three scenarios. In conse quence ASED needed to be able to handle three visual output devices simultaneously: surveillance monitor: providing all relevant information to an observer instruction monitor: displaying instructions in the state-of-the-art scenario; displaying instructions and the gamification component in the gamification scenario projector: displaying the controls and the pick indicators in all scenarios; displaying the instructions in the projection scenario Accordingly the first challenge for the implementation of ASED was finding a tech nical solution that allowed to connect all three output devices to the high performance computer required for the real-time motion detection simultaneously without produc ing glitches. Implementation: Instruction and Performance Analysis 95 Due to size restrictions USB 3.0 had to be used as a graphic data channel. The only device with the required DisplayLink-chipset available at that time (end of 2011) was the “USB 3.0 Superspeed Dual Video Adapter” by Targus which was on sale only in the USA and had to be imported. Figure 42: Instructions on a monitor (left) and as a projection (right). Once this technical issue was solved, the screen coordinates of all devices had to be synchronized. This was especially demanding with the projector because the tilt re quired by the workplace setup resulted in a keystone distortion which had to be cor rected in real-time. Since the distortion is not affine, the native components of the Swing toolkit could not apply the required transformation. Instead the OpenCV library respectively its Javaimplementation JavaCV was used. Unlike OpenCV, JavaCV does not offer a function to recognize a chess board pattern, so the generic edge detection was used. This re sulted in a semi-automatic three-step calibration process: an undistorted rectangle is projected on the workspace a picture of the recognized edges is shown on a monitor the user identifies the rectangles vertices (by clicking or touching) When the point and its projection are identified the projection matrix is calculated. The resulting CMAT (Concordance-Based Medial Axis Transform algorithm) is then transformed to deskew the image (see Equation 3). 96 Implementation: Instruction and Performance Analysis Equation 3: Transformation of distorted projection using CMAT. 6.4.2 Designer Component The ASED designer component is used when specific production sequences are mapped for the CAAS. It allows to map the steps of a production and to assign causes and effects. The visual output is also administered by this component. Each production step can hold an instruction, e.g. a picture of the product assembled up to that point and any number of interactive 3D trigger areas. These “designer ele ments” are assigned as causes and effects. A cause can be the passage of a user’s hand through a trigger area or just the exceeding of a time limit. An effect can be a sound or a navigation to the next production step. The following class hierarchy shows the designer elements implemented in ASED: Figure 43: Class hierarchy of designer elements in ASED. Implementation: Instruction and Performance Analysis 97 Although the number of causes and effects including shockwave-files, buttons and various trigger forms seems high, their implementation has been straightforward. The implementation’s main challenge was integrating all augmentation scenarios (SotA, Projection and Gamification) into one easy-to-use application. As described in the previous sub-chapter, the system had to be able to administer three visual outputs (instruction monitor, surveillance monitor and projection) because the replication of state-of-the-art assistive systems required the use of a conventional monitor. All views had to be integrated in both the designer and the runner component since all scenarios had to be configured and tested in advance. To avoid re-implement ing the panels in the Designer and the Runner component we used a SuperPanel-class inheriting from all other panels. The relations between the panels and views are shown in the following class diagram: Figure 44: Class hierarchy of panels used in ASED. On a higher level of abstraction the ASED Runner and the ASED Designer are essen tially two different graphical user interfaces (GUI) steered by the OptionPanel as in dicated in Figure 41: Architecture of the ASED software. They share the same classes and functions. Nevertheless the system requirements concerning visual outputs re sulted in a rather complex dashboard structure: 98 Implementation: Instruction and Performance Analysis 2 1 4 6 3 5 Figure 45: Screenshot of the ASED designer view. The designer component’s dashboard is segmented into six major panels labeled clockwise in Figure 45. In the center there are four large panels. 1. The top left panel is called ‘Augmented Camera Preview’ (1) and shows an image of the work area which can be switched between video and depth mode as in Figure 45. 2. The top right panel is called ‘Projector Preview’ (2) and shows the projected image (if projection is part of the scenario). In Figure 45 the preview only shows the navigation elements “next” and “previous” which are projected in all scenarios (see previous sub-chapter). 3. The steps are displayed in the panel to the right (3). 4. The panel to the bottom right is called ‘Instruction Monitor Preview’ (4) and shows the instructions assigned to the current work step. In Figure 45 the area is empty because currently the navigation element “next” is selected which is part of the global-layer. 5. The details of the selected step or the selected element within a step can be edited in the configuration area to the bottom left (5). 6. Finally to the left (6) there is a tool with different types of causes and effects to select from. The bottom left panel (5) displays information on the selected designer element and allows to edit it. This configuration area (Figure 46; 5 in Figure 45) allows to name or delete elements, and select where they are to be displayed (projection, instruction monitor, camera preview). Implementation: Instruction and Performance Analysis 99 Figure 46: Screenshot: Configuring a 3D trigger area. The sensitivity of the 3D trigger areas can be adjusted with a slider (‘Set Threshold when Box should trigger’). While the dimensions of objects in areas can be referenced (‘Reference Trigger Now’) additionally their z-range (see chap. 6.3.2 The Adjoined Spheres Approach) can be adjusted with a slider using a blue handle for the lower range and a red handle for the upper range. This is necessary for elements like the boxes: since the hand might cover the box while picking from it, the actual penetration of the trigger area by the fingers might be occluded. For fine-tuning of the calibration of trigger areas the ‘Augmented Camera Preview’ panel can be maximized to cover all four central panels (Figure 45: 1, 2, 4 and 5): Figure 47: ASED Designer with maximized “Augmented Camera Preview”. 100 Implementation: Instruction and Performance Analysis Figure 47 shows how two obstacles activate the trigger areas placed above and slightly in front of the containers. The coverage to the left is 1.304, the coverage to the right is 0.966. Although the other trigger areas also indicate coverage (0.16 and 0.164 to the left and 0.34 to the right) they remain inactive due to the threshold. The screenshot shows step 7 of the scenario projection. The instructions in Figure 47 are not assigned correctly (instruction 3 and 4 in step 7). This shows how the Designer was used in practice: work steps were copied with their calibration-values after the fine-tuning of the trigger areas. Afterwards related effects like the image of the instructions were changed subsequently. 6.4.3 Runner Component The runner component of ASED is activated whenever the CAAS is used. It handles the following tasks: Monitoring of trigger boxes by using the depth sensor. Generating an extensive automatic log file containing the times when trigger boxes are activated. Sending trigger events to the gamification component. Generating a video using the motion sensor’s RGB camera. As the list of tasks shows, the trigger box algorithm is central – it is used for interacting with the main application, for communication with the gamification component and for the logging. It strongly draws from the “adjoined spheres approach” (see chap. 6.3.2). To implement these triggers, the PrimeSense OpenNI data (see chap. 3.1 Motion Recognition) were accessed using the class VideoFrameGrabber partly developed by the author ‘init1045a’ (see chap. 9.1 Implementation Details). Based on the video and depth data the implementation of trigger boxes in Java was possible. In the runner component trigger boxes are the technical realization of the adjoined spheres model. The implementation is briefly described in the following. At first the central point of a trigger box and its distance to the corners is calculated. The closer a covered point is to the center, the greater is its value. The sum of these values results in a “filling level”. A trigger box is maximally filled if a plane parallel to the z-axis through the center of the box is covered: 101 Implementation: Instruction and Performance Analysis Equation 4: Computation of the distances to the center of a trigger box. Additional details on setting up and checking trigger boxes are provided in the sup plement section (see chap. 9.1 Implementation Details). The activities of the probands, the switching of instructions and the state of all trigger boxes can be observed in the ASED runner’s dashboard: the surveillance monitor. 1 3 2 5 4 Figure 48: Screenshot of the ASED Runner surveillance monitor. The screenshot of the ASED Runner surveillance monitor shows the ‘projector pre view’ (1, top left), the ‘instruction monitor preview’ (2, mid left – empty because in the scenario instructions are projected into the workspace) and the ‘augmented camera preview’ (3, right). To the bottom right (4) there is a time counter. The button below allows to count up processes and sequences manually – an additional fallback system (next to the video) that proved helpful when participants frequently skipped processes. The area at the bottom is used as a console and displays system information like start ing times or detected actions. 102 Implementation: Instruction and Performance Analysis Whenever a trigger box is activated a log entry is generated. As the example log entry below documents, the screenshot above was taken after the worker successfully as sembled the third working step (‘M3’) after his hand passed the trigger box of the ‘resume’ element in the navigation. Equation 5: Log entry generated by ASED Runner. While the principle of automated logging using the triggers sounds practical in theory, an undecided worker’s hand hovering over a box can generate dozens of entries – in spite of a short latency which was implemented to prevent multiple trigger activation. To clean up the resulting long log entries an analyzer software was developed (see chap. 7.3.1 Time Measurement). In some cases where assembly parts were put back in the boxes, a clear log analysis was impossible so the video documentation had to be used. Technically it was imple mented with the help of the Kinect’s video camera. The video was recorded using Xuggle, a Java-library for handling video data. In ASED it only was used to timecode the buffered images created by the class VideoFrameGrabber and to encode them into an MPEG-4 video. Implementation: Gamification Component 103 6.5 Gamification Component While the integration of motion recognition was primarily a technical task, developing the gamification component for the most part was a design task as common in games engineering. 6.5.1 Designing Flow The predominant game design question was: how can the concept Jane McGonigal calls “epic meaning” (McGonigal, 2011) at least partly be integrated into the rather repetitive work tasks in production environments? When it comes to games, most us ers talk about the “graphics” and the “sound” – so obviously these expectations had to be addressed. Less obvious, but probably more important are the game mechanics, i.e. the way a game reacts to a user’s inputs. In our case, the principle mode of inter action between the user and the CAAS has already been described in the model (see chap. 5). However, when it comes to designing the model’s concrete implementation there were no previous attempts combining real production tasks with games we could have drawn from. Based on our knowledge of the impaired users (see chap. 2.2), the indus trial scenario (see chap. 2.6 and chap. 3.4) as well as the requirements (see chap. 4), the design goal was to stay as close to the established tasks as possible. This limitation mainly is due to three reasons: The cognitive capabilities of the users are limited (see chap. 2.4.2 Cognitive Workloads) and complex stories or action elements might draw too much attention and cognitive resources away from the work process. Game-like reward structures are new to production environments, so the acceptance of obvious gaming elements like cartoon characters would re duce the acceptance rate of the decision makers in the companies (see chap. 4.3 Requirements Study) as well as the users’ acceptance rate. Pagulayan et al. see that “when comparing games to productivity software, there are principles and methods that can be successfully applied to both” (Pagulayan et al., 2012, p. 797). To put this argument further, there actually are several analogies between the real processes in manual production and gaming, e.g. building things up to become more complex and “powerful”, time limits and quality controls. Exploiting these analogies is easier and more intuitive than establishing a new frame of reference. 104 Implementation: Gamification Component The basis of the CAAS gamification component is the data generated by the Assistive Systems Experiment Designer (ASED). It allows measuring the time required for the assembly of each production component, defined as the time between two uses of the “forward”-button. As mentioned before (see chap. 6.1 System Overview) this data is not supplemented by additional emotional data, as the prototype described in this work does not allow to assess the emotional state directly, e.g. by facial detection or heart rate sensors. The basic design approach was to visually represent each work process. This was achieved by drawing from the classic tile-matching puzzle game Tetris developed by Alexey Pajitnov in 1984. Figure 49 shows a design study of the concept (left) where the work processes are color-coded and match the bricks. In the final version (right) the text element was omitted and the forms have been combined in a Tetris-like man ner to show the product’s progression. Also the colors are derived from process dura tion and not from process category. Figure 49: Design study of the gamification element (left) and final version (right). An essential requirement for gamification is that some kind of “fun” or motivation is accomplished. In the context of this work we consider this requirement as achieved, if a state of flow (see chap. 2.4.3 Flow) is reached and maintained. The following table shows how the design approach addresses the conditions for achieving the flow state. It was first proposed in the article “Industrial Playgrounds” (Korn, 2012): 105 Implementation: Gamification Component Table 6: Conditions for flow and corresponding design approach. condition being involved in an activity with a clear set of goals design approach (1) macro level: complete a (flawless) assembly sequence (2) micro level: complete the active process (=brick) as quick (=green) as possible good balance between perceived challenges and perceived skills adaptation of difficulty level based on performance task at hand must have clear and immediate feedback (1) color changes and shadowing dual-code vis ual feedback (2) sound integrates another sensory channel (1) on the micro level: “getting a stone down in time” is immediately pleasing (2) the final dissolvent of the sequence of stones appeals to the basic human desire for order and completion the activity is intrinsically rewarding In the following sub-chapter the implementation of the flow-oriented design approach is described. 6.5.2 Implementing Flow During a work process, the active brick’s color slowly changes from green to red. The duration of this color change is derived from the user’s previous process times: if for example a user completes a process in 8 seconds instead of the personal mean time of 10 seconds, this good performance will result in a dark green brick while a duration of 14 seconds would result in a yellow brick. During the first assembly sequence when no comparisons can be made, the system uses reduced process times based on meth ods-time measurement (see chap. 2.6.1 Measuring Work Processes with MTM). From the second sequence onwards the duration of corresponding processes is com pared. When a process is completed, the time difference to the last corresponding pro cess is displayed in numbers. A percentage value is generated by dividing the current duration by the mean of the recent processes. The resulting ratio is used to color-code the feedbacks and to select a short audio comment. The following table lists the feed backs used: 106 Implementation: Gamification Component Table 7: Feedbacks of the gamification component Percentage of mean duration Color Audio feedback > 200% dark red snail pace 175% - 200% light red very slow 150% - 175% orange slow 125% - 150% yellow below average 100% - 125% yellow green quite good 75% - 100% light green good 50 - 75% green very good < 50% dark green excellent For un-impaired production workers this scale would of course be stretched out too far; for impaired workers with a very high degree of performance variation between different processes, the scaling proved adequate. Figure 50: Screenshot of the instruction (left) and the gamification component (right) with shadow brick. Figure 50 shows how the worker’s mean process speed is also represented by a “shadow brick” which allows to check at any time how the current work is turning out compared to the personal average. The shadow brick’s speed can be adjusted to fit the user’s emotional state, e.g. it can be lowered if the user is in a state of negative arousal Implementation: Gamification Component 107 or increased if the user is underchallenged. Thus it serves as a guide balancing chal lenge and performance. The shadow brick was inspired by racing games, where play ers can compete against their own best rounds or those of friends or famous drivers34. The component was implemented in ActionScript 3 (AS3) and deployed as an Adobe Integrated Runtime (AIR) application. The communication with the Assistive Sys tems Experiment Designer (ASED) was realized over a datagram socket using User Datagram Protocol messages. The following code shows the implementation of the time evaluation of a process or working step: Equation 6: Process Time assessment in the gamification component. After a sequence is completed, the build-up brick rows disintegrate in an animation. Like the single processes, the sequence as a whole is compared to its predecessors. The result is a longer spoken feedback (e. g. “Congratulations – this was an excellent sequence!”) accompanied by a corresponding visual – a smiley in the adequate color and “mood” (see Figure 51). 34 A challenge between the workers would not be a technical problem, but the high variance between impaired persons gives it little practical value. Like in a single player game the user experience is essential – and multi-user games always add a sense of failure to underperforming players. 108 Implementation: Gamification Component Figure 51: The gamification component’s visual feedback makes use of smileys. The visual feedback already gives the user the important feeling of self-efficacy. How ever, in long-term operation gamification can also be used to actively influence and support the flow curve. To achieve this while giving room for the rapid performance changes common for impaired persons and while still maintaining the desired flow state as described in the model (see chap. 5), the interpreter only takes the last three assembly sequences into account. This number is an approximation based on the pre study and potentially needs further data to be established as a reliable constant (prob ably the ideal number will also vary between users). As described above, the durations of the corresponding processes of the last three sequences are compared to model the user’s emotional state and approximate the cur rent position on the flow curve. To illustrate this method, the following example shows three typical sequences (taken from the study data) with eight processes each: process duration in seconds 25 20 15 10 5 0 1 2 3 4 5 6 7 8 Figure 52: Example: Process durations of one user in eight sequences. Implementation: Gamification Component 109 The total duration of all three sequences in this example is 5.8 minutes. In the second sequence (red) the user is faster than in the first (blue). However, this changes in the last two processes 7 and 8. In sequence 3 (green) the process continue to become slower. In this example the shift on the flow curve occurs in process 7 in sequence 2 when the user starts to become slower. The change can also be observed by the trends: While usually processes tend to be completed faster within a sequence (sequence 1, blue trend line) the trend lines almost became flat in sequences 2 and 3. As described in the model (see chap. 5) the user’s hovering between flow phases is natural and necessary to sustain flow. However, an intervention can be created if the interpreter detects that the phase of arousal or control lasts too long or the trend indi cates a negative transition (e.g. from arousal to anxiety). This can be achieved by adapting the production cycle and the speed of the shadow brick. As an example an excited user can be calmed by slowing the shadow brick. This leads to more “successful” assembly processes unmistakably mirrored by green bricks and positive feedback. Thus the user’s self-confidence and motivation can be actively supported. This procedure shows that the gamification component has not been designed as a tool to increase transparency for the management or even the user – motivation and a feel ing of self-efficacy and mental stability have been give precedence over exact docu mentation. This preference and its ethical dimension are discussed in chapters 8.2 Ethical Implications: Towards Humane Work and 8.3.2 Ethical Perspectives. 110 Studies and Evaluation: Participants and Preparatory Study 7 STUDIES AND EVALUATION This chapter is based on the following publications: Korn, O., Funk, M., Abele, S., Hörz, T. & Schmidt, A. (2014) Context-aware Assis tive Systems at the Workplace. Analyzing the Effects of Projection and Gamification. In PETRA ’14 Proceedings of the 7th International Conference on PErvasive Tech nologies Related to Assistive Environments. New York, NY, USA: ACM. Korn, O., Schmidt, A., & Hörz, T. (2013). The Potentials of In-Situ-Projection for Augmented Workplaces in Production. A Study with Impaired Persons. In CHI ’13 Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Sys tems (pp. 979–984). New York, NY, USA: ACM. doi:10.1145/2468356.2468531 Korn, O., Schmidt, A., & Hörz, T. (2013). Augmented Manufacturing: A Study with Impaired Persons on Assistive Systems Using In-Situ Projection. In PETRA ’13 Pro ceedings of the 5th International Conference on PErvasive Technologies Related to Assistive Environments (pp. 21:1–21:8). New York, NY, USA: ACM. doi:10.1145/2504335.2504356 The study analyzes the effects of two augmentations (in-situ projection and gamifica tion) on the work of impaired persons in production environments. 7.1 Participants and Preparatory Study The field study was conducted at the Beschützende Werkstätte Heilbronn (BWH), a German sheltered work organization. The persons working there suffer from various kinds of cognitive and motoric impairments – ranging from stroke-related problems and epilepsies to Down syndrome. To ensure ethical compliance, we aligned the study design to the “PD-Net Approach to Supporting Ethics Compliance” (Langheinrich, Schmidt, Davies, & José, 2013) which was primarily designed for research involving public displays. Although it was not possible to follow the proposed method to the letter and establish an independent ethical advisory board, the BWH is already split in a profit and a non profit unit. The governing board of both is chaired by a priest. It is organized as a registered society and has incorporated the Corporate Governance Codex of the Diakonisches Werk, a huge Protestant social welfare association with over 450.000 employees and about 700.000 volunteers. All materials used in the study as well as the documented procedures have been acknowledged by the governing board. Studies and Evaluation: Participants and Preparatory Study 111 While all persons at the BWH are able to work, their competence levels vary strongly according to the kind of impairment. Although (almost) all of them are able to eat on their own or express thoughts in simple words, some would not be able to count to ten while others could perform algorithmic operations like multiplying. A similar varia tion occurs in assembly skills. While the augmented context-aware assistive systems (CAAS) are being designed to be “universal” and assist as many persons as possible, the competence level of many impaired persons at the BWH would not suffice to per form assembly tasks at all. So the first task was finding a population of test subjects suited for simple assembly tasks in a preparatory study. To avoid a pre-assessment of all impaired persons, the participants were pre-selected based on their wage-level number. This number is generated by the sheltered work organization based on an extensive analysis of an impaired person’s skills. It includes a physical and a cognitive rating as well as psychological factors like work motivation and stress resistance. To find a suitable range to start with, we interviewed the work instructors and broadened the recommended wage level range to ensure no potential subjects were lost. A range between 55 and 125 resulted in 135 potential test subjects. To uphold the established ethical standards of data gathering (Langheinrich et al., 2013; Runeson & Höst, 2008) the test subjects or their legal guardians were contacted by the sheltered work organization’s board and asked to sign a written consent to par ticipate in the experiment. This form guaranteed anonymity and explained the benefits arising from augmented or enhanced assistive systems in simple language. Due to that direct access and the beneficial aim of the study, the return quota was very high: Al together 100 of the 135 impaired persons contacted (74%) or their respective guardi ans consented to participate in the experiments. The pre-study was conducted to match the complexity of the assembly task and the cognitive skills of the impaired workers. The participants were asked to assemble Lego objects of varying complexity reaching from a simple asymmetrical bridge to a rather complex vehicle (see Figure 53). Figure 53: Lego objects with various degrees of complexity in the pre-study. 112 Studies and Evaluation: Participants and Preparatory Study Technically speaking, the pre-study’s goal was finding the upper and the lower bound of the wage level range. The boundary conditions were defined as follows: lower bound: the inability to assemble an object consisting of four parts upper bound: the ability to assemble an object consisting of four parts from memory at the fourth repetition The pre-study showed that the wage level range required to perform simple assembly work usually lies between 70 and 125. This requirement narrowed down the number potential test subjects from 100 to 81. The resulting test population was divided into three groups or scenarios35: (1) State-of-the-Art: no augmentation (2) Projection: Augmentation by in-situ projection of instructions (3) Gamification: Augmentation by gamification The participants were distributed over the three groups; to sustain an average wagelevel index among the groups, the selection process was designed as follows: sort participants based on their wage-level number sequentially add a participant to the group 1, 2 and 3 if one group’s mean wage-level index differs by more than 2 points, take the participant that is best suited to equalize the wage-level scores Figure 54 shows a histogram of the wage level distribution in the three groups. Under ideal conditions the bars would be of equal length in all classes, in practice the limited number of test subjects (60 out of 81) and the “re-randomization” in the course of the experiments due to illnesses and dropouts resulted in more varied distributions. 5 4 3 2 1 0 71‐80 81‐90 91‐100 SotA Projection 101‐110 111‐120 121‐130 Gamification Figure 54: Histogram of the three test populations. 35 The terms SotA, Projection and Gamification written with a capital letter refer to the corresponding scenarios or test populations. 113 Studies and Evaluation: Participants and Preparatory Study The mean wage level (Figure 55) for the SotA test population is = ̅ݔ100.7 with a stand ard deviation (SD) of 13.4, for the scenario Projection it is = ̅ݔ100.2 (SD=19.0) and for the scenario Gamification it is = ̅ݔ101.3 (SD=17.3) – so the variance was greater in the two groups working with the augmented system. However, the differences between the three test populations altogether are minimal and not significant (p>0.978 in an ANOVA over all scenarios). 120 60 80 age in years wage level 100 60 40 40 20 20 0 0 SotA Projection Gamification SotA Projection Gamification Figure 55: Test populations’ mean wage-level (left) and age (right) with SD. With regard to age the mean for SotA subjects is = ̅ݔ42.5 (SD=10.8), with the Gami fication subjects being slightly older with = ̅ݔ43.5 (SD=10.8) and the Projection sub jects being slightly younger with = ̅ݔ39.8 (SD=11.5). Again the differences are very small and not significant (p>0.553 in an ANOVA over all scenarios). Summing up the analysis shows that of the differences between the three test popula tion in the independent variables wage level and age are insignificant so the observed effects can be attributed to the augmentations. 114 Studies and Evaluation: Procedure 7.2 Procedure Each experiment consisted of four phases: 1. initial interview and questionnaire part 1 2. short introduction to the system used in the specific scenario 3. assembly phase (eight sequences) 4. questionnaire part 2 and final interview For phase 1, 3 and 4 the test persons’ attitude towards the specific system used and the resulting work experience was measured with a questionnaire. During the assem bly phase the process times were measured using the CAAS prototype and the ASED software (see chap. 6 Implementation). Assembly errors were documented on photos. The procedure is described in more detail in the following sub-chapters. Qualitative findings from the interviews are described in the results section (see chap. 0). 7.2.1 Questionnaire Design The questionnaire is an adapted version of the established system usability scale (SUS) (Lewis & Sauro, 2009). We also drew from the well-described AttrakDiff (Has senzahl, Burmester, & Koller, 2008). Based on our experiences with impaired users we reduced the number of items on the Likert-scale from seven to five. The questionnaire consists of a generic part which is identical in all scenarios and a scenario-specific part. As a result, the SotA version of the questionnaire contains 19 questions, the Projection version contains 22 questions and the gamification version contains 24 questions. Furthermore the questionnaire is divided in a pre and a post experiment section. The generic questions aim to determine the physical and emotional state of the test subjects before the experiment. A second aim was determining if the proband is ex perienced with computers or computer games, since measuring performance in a computer-based system always implies the risk of measuring computer affinity and thus creating reliable but non-valid results. The following five generic questions were posed before the assembly phase: 1. How are you today? 2. Are you nervous? 3. Do you look forward to the experiment? 4. How many hours per week do you work at the computer? 5. How many hours per week do you play computer or console games? Studies and Evaluation: Procedure 115 The next 11 generic assertions were presented after the assembly phase: 6. The help device36 was easy. 7. I would use the help device again. 8. I would recommend the help device to my colleagues. 9. I need somebody’s help to use the help device. 10. The help device was unnecessary complicated. 11. The indicator for picks was easy. 12. I found it tiring to confirm each step with a button. 13. Working with the help device can be learned quickly. 14. I found the work with the help device exhausting. 15. I could use the help device proficiently. 16. I had to learn a lot before I could work with the help device. The assertions in this list are deliberately redundant. Also contradictory items are used (e.g. items 6 and 10 or items 15 and 16). The aim was to detect the tendency to “yes”, i.e. probands always answering in the affirmative. The order of the items in the ques tionnaires was different from the list presented here – the items were scrambled to make the redundancies and contradictions less obvious. The original questionnaires are provided in the supplement section (see chap. 9.2 Questionnaires in Detail). Item 11 refers to the number projected in front of the box to pick from in all scenarios which emulates an industry standard pick-by-light system (see chap. 3.4.2 Assistive Systems Using Projection). The following table shows the specific questions for each scenario posed after the assembly phase. 36 We tried to use “simple language” whenever possible and avoided the term “assis tive system”. However, all questions were posed in a dialogue and could be rephrased if the test subject had difficulties understanding them. 116 Studies and Evaluation: Procedure Table 8: Scenario-specific questions posed after the assembly phase. SotA Projection Gamification I found the instruction on the monitor complicated. I found the instruction left of me on the table compli cated. I found the instruction on the monitor complicated. I found the instruction on the monitor easy. I found the instruction left of me on the table easy. I found the instruction on the monitor easy. I found it tiring to always check the instruction on the monitor above me. I liked having the instruc tion directly to my left on the table. I found it tiring to always check the instruction on the monitor above me. - The instruction left of me on the table disturbed my work. - - The model in the center of the workspace helped me. - - The model in the center of the workspace bothered me. - - - I liked the game on the monitor. - - The game on the monitor bothered me. - - The comments on my work motivated me. - - The comments on my work disturbed me. - - I would like to always have the game during my work. Studies and Evaluation: Procedure 117 7.2.2 Assembly Task The test subjects were asked to assemble eight identical car undercarriages (no varia tions) using Lego bricks. Each of these eight assembly sequences consists of eight assembly processes (short: “steps”) so a perfect experiment consisted of 64 successful assembly steps. During each steps a corresponding instruction was shown (see Figure 56) either on the monitor or directly in the workspace. After completing a step the test subjects had to use the permanently projected navigation element (see chap. 6.4.1 Visual Output and Projection) to go to the next step by touching the green “resume” button or to go back to the previous step by touching the red “back” button. Figure 56: Instructions in the scenarios Gamification (left) and Projection (right). While the process of picking the right parts for the assembly process at hand was identical in all three scenarios (the boxes to pick from were marked by a number), the difference between the SotA and the Projection scenario was the way instructions are presented to test subjects: while the scenarios SotA and Gamification use a monitor, in the Projection scenario the work-relevant information is projected directly into the user’s workspace. In addition to the instructions a 1:1 model of the correct result of the assembly is projected directly into the center of the workspace; also the active step is marked by a green arrow (see Figure 56). 118 Studies and Evaluation: Procedure The instructions themselves remained identical in all three experiments: for each of the eight assembly steps a product image before and after that specific assembly step is shown. The instructions used are displayed in the table below. Table 9: Instructions for the eight assembly steps. Process 1-4 Process 5-8 The red steps (M1 and M4) were the ones with deliberately designed challenges like asymmetries (see chap. 7.3.2 Quality measurement). Studies and Evaluation: Apparatus 119 7.3 Apparatus The physical features of the prototypical CAAS, i.e. the design of the assembly work place and the integration of the sensors and the projector, have already been described (see chap 6.2). In the scenario State-of-the-Art (SotA) and the scenario Gamification just the monitor is used. In the latter both the instruction and the gamification element were shown on the monitor. While it would have been technically possible to project them into the workspace and thus combine projection and gamification, this combined scenario would not have allowed to analyze the effects of the two augmentation sep arately. However, the navigation and the indicator for the picks were always projected to avoid discrepancies between the users’ performance related to minor interface is sues (see chap. 6.4.1 Visual Output and Projection). 7.3.1 Time Measurement All three scenarios were documented using the ASED software (see chap. 6.4 Instruc tion and Performance Analysis). Each passage through a trigger area generated an entry into an XML-file as illustrated by the example in Equation 7: Equation 7: Log entry generated by ASED Runner. During the field study several test subjects forgot to confirm processes or mixed up process steps, so log files were incomplete or out of sync. Some participants also un decidedly hovered their hands over the component boxes, generating dozens of entries within a few seconds while others used the “next” and “previous” buttons excessively. As a result some log files contained several hundred entries where potentially (in a perfect experiment) 128 entries would suffice. 120 Studies and Evaluation: Apparatus Since we had already encountered these problems in the pre-study, we addressed them in three ways: the software captured a video of the workspace the software allowed the observer to manually log process times in addition to the automatic logs as a software-independent fallback we used a camcorder to record each experiment Although the video documentations allowed the manual supplementing of flawed logs this was a “final resort”. To avoid checking all log files manually, the “Logfile Ana lyzer” was developed for the automatic parsing and cleanup of extensive XML logs. It was implemented in PHP and MySQL using the Zend Framework and jQuery. The analyzer parses both the automatic and the manual log file using the extended parser class and stores the data into a parser object as associative arrays. Once the data is parsed it is shown in a list view with the automatic data on the left and the manual data on the right. Discrepancies are highlighted in red color. In this view functions are used to optimize the log-files: e.g. ‘del’ (deletes working step and adds duration to the previous step), ‘moveRight’ (inserts automatic log step into manual log step and recalculates aggregation) or ‘use’ (uses the duration stored in the manual log instead of the automatic log duration). Figure 57 and Figure 58 show a logfile before and after optimization. Figure 57: Logfile Analyzer list view before optimization. Studies and Evaluation: Apparatus 121 Figure 58: Logfile Analyzer list view after optimization. In this case the resolution required the following deductions: Firstly in step 28 the process declaration P2 follows a P3-process and precedes a P4-process. Secondly step 28 has a duration time of 0.0 s. The obvious solution is removing step 28. 7.3.2 Quality measurement The eight steps required for the assembly sequence used in the experiment vary in complexity. Based on our experience from the pre-study we deliberately included two major potential sources of error highlighted red in Table 9 Instructions for the eight assembly steps: The first step requires abandoning the desire for symmetry because the first Lego cube is not aligned with the base plate. The fourth step requires to attach the 4x1 Lego plate at the correct side of the middle plate. Unlike time measurement the analysis of errors had to be performed manually. Each of the 480 assembled products or 3840 steps in the three experiments was analyzed individually. In this analysis the outcome was looked at favorably – so a mistake in step one (e.g. aligning the cube with the plate) that was neutralized in step two (assembling the cube as required in step one) was not counted as two but as zero mistakes. From a produc tion management view such a change in sequence would be interpreted as two mis takes, even if the result was correct. 122 Studies and Evaluation: Apparatus Our more favorable approach is due to the fact that even by interpreting videos with facial expressions it is impossible to ascertain if a correct assembly is the result of chance or the test subject’s deliberate correction of a previous mistake. Figure 59: Undercarriage assembled correctly (left) and with 100% error rate (right). When looking at the error rates it is important to consider that a 100% error rate (see Figure 59) is very rare. Although it did appear a few times, an assembly with an error rate above 80% requires a violation of several “common sense concepts” – e.g. wheels have to be assembled on top of the undercarriage or in a way that the vehicle would drive in circles or sideward. Studies and Evaluation: Data 123 7.4 Data In this chapter we will portray the data “as it is” using only descriptive statistics (means). However even this data is aggregated – for example we will look at sequence completion times rather than process completion times (there are eight processes in each sequence, i.e. 64 processes to analyze for each user). The complete data is pro vided in the supplement section (see chap. 9.3 Study Result Details). In the ensuing sub-chapters we will look deeper and use inferential statistics like the Student’s t-test and ANOVA (analysis of variance). In the following, the subsequent data graphs are provided: SotA population o Sequence completion time for all participants o Error rate for all participants Projection population o Sequence completion time for all participants o Error rate for all participants Gamification population o Sequence completion time for all participants o Error rate for all participants Complete Population o Histogram of average assembly time o Histogram of average error rate 124 Studies and Evaluation: Data 600 subject 1 time in seconds to complete an assembly sequence subject 2 subject 3 subject 4 500 subject 5 subject 6 subject 7 400 subject 8 subject 9 subject 10 300 subject 11 subject 12 subject 13 subject 14 200 subject 15 subject 16 subject 17 100 subject 18 subject 19 subject 20 0 1 2 3 4 5 6 7 8 Figure 60: Sequence completion time of the SotA population. 800% 700% subject 20 subject 19 subject 18 error rate in the assembly sequences 600% subject 17 subject 16 subject 15 500% 0% 38% 50% 400% 38% 50% subject 14 25% 13% 0% 13% 0% 50% 63% 25% 25% 0% 38% 13% 25% 13% 63% subject 13 13% 50% 13% 13% 25% 0% 50% 63% 300% 63% 25% 25% 0% 38% 200% 38% 38% 13% 0% 38% 13% 100% 63% 0% 13% 0% 38% 0% 13% 1 38% 25% 25% 0% 38% 13% 50% 0% 13% 38% 0% 13% 0% 38% 13% 63% 63% 13% 0% 38% 25% 13% 2 63% 25% 13% 25% 0% 3 subject 12 13% 25% 13% 75% 75% 25% 13% 25% 13% 0% 38% 0% 38% 13% 63% 25% 13% 13% 38% 0% 13% 0% 13% 63% 13% 0% 13% 13% 25% 13% 50% 0% 4 50% 0% 5 63% 63% 50% 25% 0% 13% 25% 13% 13% 0% 38% 13% 63% 25% 25% 13% 25% 25% 13% 0% 38% 13% 63% 63% 0% 25% subject 11 subject 10 subject 9 subject 8 subject 7 subject 6 subject 5 38% 0% 13% 0% 13% subject 4 63% subject 2 13% 0% 13% 25% 25% 13% 0% 13% 0% 25% 13% 0% 13% 25% 25% 6 7 8 Figure 61: Error rate per sequence of the SotA population. subject 3 subject 1 125 Studies and Evaluation: Data 600 subject 1 time in seconds to complete an assembly sequence subject 2 subject 3 subject 4 500 subject 5 subject 6 subject 7 400 subject 8 subject 9 subject 10 300 subject 11 subject 12 subject 13 subject 14 200 subject 15 subject 16 subject 17 100 subject 18 subject 19 subject 20 0 1 2 3 4 5 6 7 8 Figure 62: Sequence completion time of the Projection population. 800% 88% 700% subject 20 25% subject 19 63% subject 18 error rate in the assembly sequences 600% 100% 88% subject 17 38% subject 16 100% 38% 500% 0% 25% 13% 25% 0% 13% 0% 50% 100% 13% 400% 100% 63% 0% 13% 13% 13% 13% 25% 0% 13% 0% 38% 0% 300% 63% 50% 200% 63% 100% 63% 13% 13% 38% 100% 38% 25% 13% 63% 13% 0% 13% 13% 25% 13% 0% 13% 0% 13% 0% 13% 63% 88% 13% 13% 13% 0% 13% 0% 13% 0% 13% 13% 13% 13% 0% 13% 0% 13% 0% 13% 63% 63% 13% 0% 13% 0% 13% 0% 100% 13% 100% 100% 38% 63% subject 13 13% 75% 13% subject 12 75% 13% 13% 13% 13% 0% 13% 0% 13% 13% 13% 75% 13% 13% 13% 0% 13% 0% 13% 0% 13% 63% 63% 25% 0% 0% 13% 13% 0% 13% 0% 13% 13% 13% 63% 25% 0% 25% 0% 100% 100% 100% 88% 100% 75% 0% 63% subject 11 subject 10 subject 9 subject 8 subject 7 subject 6 subject 5 subject 4 subject 3 subject 2 88% 75% subject 15 subject 14 13% 63% 13% 13% 38% 13% 13% 25% 1 2 3 4 63% 50% 5 subject 1 63% 25% 6 63% 50% 7 25% 8 Figure 63: Error rate per sequence of the Projection population. 126 Studies and Evaluation: Data 600 subject 1 subject 2 time in seconds to complete an assembly sequence 500 subject 3 subject 4 subject 5 subject 6 400 subject 7 subject 8 subject 9 subject 10 300 subject 11 subject 12 subject 13 200 subject 14 subject 15 subject 16 subject 17 100 subject 18 subject 19 subject 20 0 1 2 3 4 5 6 7 8 Figure 64: Sequence completion time of the Gamification population. 800% 700% 50% 25% 0% 38% 25% 0% 13% error rate in the assembly sequences 600% 50% 500% 25% 0% 50% 13% 13% 38% 13% 300% 200% 50% 0% 13% 13% 38% 13% 25% 25% 13% 13% 0% 13% 13% 38% 13% 25% 25% 25% 0% 13% 25% 50% 13% 38% 25% 63% 100% 88% 63% 0% 13% 0% 1 50% 0% 13% 13% 38% 38% 0% 13% 13% 25% 50% 50% 50% 25% 13% 13% 38% 50% 63% 50% 0% 13% 0% 38% 63% 38% 13% 25% 13% 25% 25% 13% 2 3 4 13% 0% 5 subject 16 subject 14 subject 13 subject 12 subject 11 13% 25% 13% 13% subject 10 50% subject 9 subject 8 subject 7 25% 13% 38% 38% 0% 25% subject 5 100% 100% 63% 63% 13% 25% 25% 13% 25% 13% 38% 0% 13% 6 7 8 38% 25% 100% subject 6 subject 4 subject 3 subject 2 63% 63% subject 17 63% 100% 63% subject 18 subject 15 63% 63% 38% 88% 63% 25% 25% 25% 25% 88% 50% 38% 63% 75% 13% 50% 50% 25% 38% 0% 13% 63% 63% 38% subject 19 50% 50% 38% 50% 50% 50% 75% 63% 400% 63% 63% 50% subject 20 38% 50% 38% 50% 63% 63% 38% 25% 38% 0% 13% subject 1 Figure 65: Error rate per sequence of the Gamification population. 127 Studies and Evaluation: Data 10 number of test subjects 9 8 7 SotA 6 Projection 5 Gamification 4 Poly. (SotA) 3 2 Poly. (Projection) 1 Poly. (Gamification) 0 0‐50 50‐100 100‐150 150‐200 200‐250 250‐300 300‐350 350‐400 mean time for one sequence in seconds Figure 66: Histogram of average assembly time (eight sequences). number of test subjects 12 10 8 SotA 6 Projection 4 Gamification Poly. (SotA) 2 Poly. (Projection) 0 Poly. (Gamification) mean error rates categrized Figure 67: Histogram of average error rate (eight sequences). Already by looking at the data as presented here, several observations can be made: In all scenarios there is a high variance between the test subjects. In all scenarios there is one subject completely divergent when it comes to time measurements (SotA: 9, Projection: 1, Gamification: 19). In all scenarios there is some kind of learning effect concerning speed: subsequent sequences tend to be completed faster. In all scenarios speed is roughly normally distributed among the subjects. The learning effect does not seem to imply errors: they either roughly remain constant (SotA) or even go up (Projection and Gamification). In all scenarios errors are not normally distributed among the subjects. More errors occur in the augmented scenarios. 128 Studies and Evaluation: Experiment Results and Discussion 7.5 Experiment Results and Discussion The following analysis is based on our previous work on the effectiveness of augmen tations for workspaces (Korn, Funk, Abele, Hörz, & Schmidt, 2014; Korn, Schmidt, & Hörz, 2013). 7.5.1 Overall Results While we looked at the whole test population in the analysis of the independent vari ables wage level and age (see chap 7.1 Participants and Preparatory Study), the de pendent variables in the experiments are process time and error rate. We used both analysis of variance (ANOVA) and Student’s t-test to evaluate the results. Because of the high variance between single participants already p-values below 0.1 (marginally significant) are considered significant results. Although the prototype of the CAAS used for this study could not provide productrelated feedback on errors (see chapter 4.2.2 Total Quality and chapter 6.3.1 Technical Restrictions and Solution) the effects of the augmentations on quality are an important subject of the study. Mostly two p-values will be discussed because the difference between the state-of-the-art (SotA) scenario and one of the two augmented scenarios is relevant – i.e. the difference between SotA and Projection or between SotA and Gamification. The following table shows the results on a macro level and also lists the standard deviations (SD). The delta (∆) refers to the SotA values. Table 10: Means and standard deviations of sequence durations and errors. Variable SotA Projection Gamification Mean production time = ̅ݔ25.6 min = ̅ݔ23.6 min ∆ = -7.8% = ̅ݔ22.4 min ∆ = -12.5% SD = 9.0 min SD = 10.4 min ∆ = 15.5% SD = 6.9 min ∆ = 23.3% = ̅ݔ22.6% = ̅ݔ29.1% ∆ = 28.7% = ̅ݔ33.1% ∆ = 46.5% SD = 17.5% SD = 27.3% ∆ = 56.0% SD = 22.1% ∆ = 26.3% Mean error rate The mean duration is reduced by both augmentations with Gamification having a greater impact than Projection. On the other hand the error rates rise as a result of the augmentations, so potentially a speed-accuracy trade-off has occurred. The high var iance among the participants results in high standard deviations in all scenarios. Studies and Evaluation: Experiment Results and Discussion 129 7.5.2 Analysis of Time As Figure 68 shows, the differences in time needed for the assembly sequences were comparatively small with large variations: mean production time in minutes 35,00 30,00 25,00 20,00 15,00 10,00 5,00 0,00 SotA Projection Gamification Figure 68: Mean sequence durations with SD of the test populations. If all 40 users are taken into account, the h1-hypothesis that CAAS augmented by projection will make the users faster (with the h0-hypothesis assuming that there is no significant difference) cannot be confirmed. A t-test shows that the average time re duction of 7.8% is not enough to be statistically significant (p>0.259 one-sided), thus h0 is maintained. The analogous h1-hypothesis for the augmentation by gamification also cannot be confirmed – the average time reduction of 12.5% is not enough to make the difference between the two scenarios statistically significant (p>0.110 one-sided). However, the comparatively low p-values indicate that research with larger or more homogenous groups or an analysis of aggregated data probably will show that gamification does improve production speed. 130 Studies and Evaluation: Experiment Results and Discussion sequence completion time (s) When looking at the graphs in Figure 69 which show how the group means developed over the eight production sequences, there clearly are relevant differences between the test populations: 400 350 300 250 200 150 100 s1 s2 s3 s4 s5 s6 s7 s8 SotA 348 227 197 179 170 150 131 133 Projection 313 221 177 175 146 133 135 115 Gamification 290 202 169 152 137 137 133 125 Figure 69: Development of mean sequence completion times over eight sequences. As expected the sequence times in all three test populations drop – very probably due to learning effects and the users’ adjustment to the setup and the system setup used. In this aggregated perspective there are no significant differences between SotA and Projection: the h1-hypothesis that a CAAS augmented by projection will significantly reduce the mean sequence completion times is falsified (p>0.200 one-sided). However, in the aggregated perspective the h1-hypothesis that CAAS augmented by gamification will significantly reduce the mean sequence completion times is con firmed (marginally significant with p<0.064 one-sided). 131 Studies and Evaluation: Experiment Results and Discussion process completion time (s) Since the processes within the eight sequences do not change (i.e. there are no varia tions) it is possible to compare the development of process times: the mean times of all processes of one specific assembly. In opposition to the horizontal perspective of the sequence durations, the process durations add a “vertical perspective”. 400 350 300 250 200 150 100 p1 p2 p3 p4 p5 p6 p7 p8 SotA 258 162 185 185 185 154 182 225 Projection 255 145 183 189 159 146 177 162 Gamification 223 151 162 161 163 153 173 158 Figure 70: Development of mean process completion times. This form data visualization shows that the lines representing the process times are mostly parallel (see Figure 70), so no process seemed to benefit especially. An interesting observation in this vertical perspective is that the subjects in the SotA scenario needed significantly more time for the eighth process. The h0-hypothesis that the eighth process will be completed faster than the previous process can be rejected (p<0.008 one-sided when compared with Projection; p<0.005 one-sided when com pared with Gamification). While the low p-values indicate that this observation is no random effect, a reliable explanation would require more data. We can only speculate that the stimulative effect of both augmented scenarios countered a natural speed reduction when the product is “almost finished”. 132 Studies and Evaluation: Experiment Results and Discussion 7.5.3 Analysis of Errors With respect to the errors made during the eight assembly sequences, the difference between the three scenarios were much more obvious than with production speed (see Figure 71). Again there were strong variations. 60,0% 50,0% error rate 40,0% 30,0% 20,0% 10,0% 0,0% SotA Projection Gamification Figure 71: Mean error rates with SD of the test populations. When looking at the data of the 40 users, the h1-hypothesis is that CAAS augmented by projection which do not provide feedback on product-specific errors will increase the number of errors. The corresponding h0-hypothesis assumes that there is no sig nificant difference. The results show that the h1-hypothesis cannot be confirmed and h0 is maintained: the mean increase in errors (6.1% absolute and 28.7% relative) is not enough to be statistically significant (p>0.186 one-sided). While this result is “good” for CAAS, the high variance shows that further research with larger or more homog enous groups is required. The parallel h1-hypothesis that CAAS augmented by gamification not providing feed back on product-specific errors will increase the error rate can be confirmed. In spite of the high variance the average increase in errors (10.5% absolute and 46.5% relative) is statistically significant (p<0.052 one-sided) so h0 is falsified. Obviously the users’ motivational gain from gamification was transformed into speed. And since no feed back on quality was given, a speed-accuracy trade-off occurred. The analysis of mean error rates follows the schema introduced with the analysis of time: the horizontal perspective focusing on the errors made successively over eight sequences is complemented by the vertical perspective focusing on the errors made in each one of the eight processes. 133 Studies and Evaluation: Experiment Results and Discussion 40% sequence error rates 35% 30% 25% 20% 15% 10% s1 s2 s3 s4 s5 s6 s7 s8 SotA 24% 24% 21% 23% 23% 23% 22% 20% Projection 39% 28% 24% 24% 31% 31% 29% 27% Gamification 29% 29% 31% 34% 36% 37% 35% 34% Figure 72: Development of mean sequence error rates. The SotA line in Figure 72 shows that the error rate almost stays constant. This means that there was no learning effect. Although this can partially be attributed to the lack of quality-related feedback it is important to note that eight repeated assemblies of an identical product did not result in decreased error rates. In this aggregated perspective with reduced variance (which does not compare the error rates of all 40 users but the users’ average error rates for the eight sequences) the differences between the scenarios are more accentuated. The h0-hypothesis assum ing that there is no significant difference in mean sequence error rates is falsified for Projection (p<0.003 one-sided), so as expected from the previous findings this aug mentation produces significantly more errors. However, the development of the graph with a quick reduction of errors until the fourth sequence and a sudden raise in the fifth cannot be explained by quality-related versus speed-related feedback and demands a more thorough analysis (see chap. 7.5.4 Sub-Populations). The h0-hypothesis is also (again) falsified for Gamification, although in the aggre gated view the difference becomes much more evident (p<0.00001 one-sided). The upwards trend of the error line in Gamification can be explained by the implementa tion specifics: since the users only received feedback on the speed of production, fast assembly became their focus while quality was neglected. Thus the raising error line is very probably the result of a speed-accuracy trade-off induced by the positive feed back and potentially a flow state. 134 Studies and Evaluation: Experiment Results and Discussion 70% process error rates 60% 50% 40% 30% 20% 10% 0% p1 p2 p3 p4 p5 p6 p7 p8 SotA 56% 3% 35% 44% 27% 14% 0% 1% Projection 67% 15% 24% 39% 32% 29% 14% 13% Gamification 56% 29% 39% 51% 34% 46% 7% 5% Figure 73: Development of mean process error rates. Figure 73 shows that all participants had problems at the start of a new sequence (pro cess 1). While processes 2, 7 and 8 usually were acceptable (error rates below 15%) the development within the processes 3, 4, 5 and 6 is interesting. Over all scenarios the error trend consistently goes down from process 5 to 8, the graphs run in parallel. However, there is one exception in Gamification where process 6 even produces a substantial 12% more errors than process 5. This process is looked at in more detail: process 6 error rates 70% 60% 50% 40% 30% 20% 10% 0% s2 s3 20% 5% 50% 25% Gamification p6 45% 35% SotA p6 Projection p6 s1 s4 s5 s6 s7 s8 25% 5% 20% 10% 15% 15% 20% 25% 25% 30% 30% 25% 40% 40% 50% 55% 55% 55% Figure 74: Development of mean error rates of process 6 in eight sequences. 135 Studies and Evaluation: Experiment Results and Discussion The h0-hypothesis in this case is that process 6 in Projection and Gamification does not differ significantly from process 6 in SotA. While in Projection the difference to SotA is barely significant (p<0.005 two-sided), with Gamification the statistical sig nificance reaches astronomical heights (p< 0.000001 two-sided). However, it remains unclear why the sixth process in Gamification consistently differs from the seven other processes which develop congruently to the other scenarios. 7.5.4 Sub-Populations The development of the projection data with the quick reduction of errors until the fourth sequence and the sudden raise errors in the fifth sequence demanded a more thorough analysis. The basic idea was to compare two sub-populations: the faster users completing the sequences faster than the mean time and the slower users requiring more than the mean time. 35,00 60,0% 30,00 50,0% 25,00 40,0% error rate mean production time in minutes The analysis is based on previous work on the effectiveness of in-situ projection as an augmentation for workspaces (Korn et al., 2013). 20,00 15,00 30,0% 20,0% 10,00 10,0% 5,00 0,00 0,0% SotA Proj. Game SotA Proj. Game Figure 75: Production times (left) and errors (right) in the complete population (left column) and the faster group (right column). Figure 75 shows that in the Gamification scenario the faster users (left bar) did not reduce errors so this scenario is not part of the analysis. In SotA and Projection there are twelve test subjects performing faster than average (referred to with ‘+’). The mean sequence in SotA+ took = ̅ݔ151 s (SD = 36.6 s). In Projection+ it took = ̅ݔ129 s (SD = 31.9 s): a time reduction of 22 seconds or 14.5%. The hypothesis h1 that projection increases the speed of above average workers can be accepted (p<0.061 one-sided). Obviously the over-performers benefited more from the in-situ projection of instructions than the group as a whole. 136 Studies and Evaluation: Experiment Results and Discussion The development of the product quality in the faster subgroup as illustrated in Figure 76 is even more interesting: 40% 35% 34% error rate 30% 25% 28% 20% 29% 27% 14% 15% 23% 11% 29% 26% 30% 15% 14% 24% 9% SotA+ Proj+ 11% 9% 10% Expon. (SotA+) Expon. (Proj+) 5% 0% 0 1 2 3 4 5 6 7 8 production sequences Figure 76: Temporal development of mean error rates in the subgroups. The SotA+ group’s mean error rate is = ̅ݔ27.1% (SD = 20.5%). Thus the error rate of the fast workers is 4.5% higher than the complete population’s average which indi cates that the speed gain partly is the result of more careless assembly: a speed-accu racy trade-off. Also there is no quality improvement in the SotA+ population as the trend moves sideward. The Projection+ group differs strongly: here the mean error rate is only = ̅ݔ14.7% with a comparatively low standard deviation (SD = 15.5%). Given that the complete Projection population’s mean error rate is 29.1% the faster users made only about a third of the mistakes of the slower users. Compared to the SotA+ population there is a 45.8% error reduction (relative value based on the 27.1% error rate in SotA+ and the absolute reduction of 12.4%). Thus the hypothesis h1 that projection decreases the error rate of above average workers can be accepted (p<0.053 one-sided). This is es pecially important because these workers – like all the others in the experiments – did not receive feedback on product-specific errors. Probably they made good use of the 1:1 model projected in the center of the workspace (see chap. 7.2.2 Assembly Task). Finally there is a clear trend towards making less errors in this group, so augmenting a workplace with projection also seems to trigger learning and thus continually im proves quality. 137 Studies and Evaluation: Questionnaire Results and Discussion 7.6 Questionnaire Results and Discussion The questionnaire was designed to collect objective quantitative data on how contextaware assistive systems (CAAS) in production environments are experienced by their users (see chap. 7.2.1 Questionnaire Design). Each group or scenario was comprised of 20 participants testing a specific version of the CAAS: SotA (no augmentation), Projection and Gamification. To analyze the data we used analysis of variance (ANOVA) and Student’s t-test. 7.6.1 Pre-Experiment Results In the first questions the participants’ general condition and their experience with computers was checked. Table 11: Questionnaire item 1: General condition 1. How are you today? SotA Proj. Gamif. ( ̅ݔSD) 2.2 (0.8) 2.0 (0.5) 2.1 (0.2) 1: very good, 2: good, 3: normal, 4: not good, 5: bad Although there is some difference in variance, there was no significant difference in the condition (p>0.47) before the participants started the experiment. Table 12: Questionnaire item 2: Nervousness 2. Are you nervous? SotA Proj. Gamif. ( ̅ݔSD) 2.2 (1.1) 2.7 (0.7) 2.6 (0.6) 1: not at all, 2: very little, 3: normally, 4: distinctly, 5: very much Although there is some difference in variance and the participants in Projection and Gamification were more nervous than in SotA, there was no significant difference in nervousness (p>0.206) before the participants started the experiment. 138 Studies and Evaluation: Questionnaire Results and Discussion Table 13: Questionnaire item 3: Anticipation 3. Do you look forward to the experiment? SotA Proj. Gamif. ( ̅ݔSD) 3.7 (0.8) 3.4 (0.7) 3.5 (0.5) 1: not at all, 2: very little, 3: normally, 4: distinctly, 5: very much There is little difference in variance, the participants are in positive anticipation of the experiment. Of sixty participants only one selected “very little”. There also was no significant difference in anticipation (p>0.328) before the participants started the ex periment. Table 14: Questionnaire item 4: Computer experience 4. How many hours per week do you work at the computer? SotA Proj. Gamif. ( ̅ݔSD) 1.9 (1.1) 1.7 (0.9) 1.7 (1.0) 1: not at all, 2: < 2 hours, 3: 2-5 hours, 4: 5-10 hours, 5: > 10 hours With little difference in variance, most participants have little to do with computers – mostly they encounter them at educational gaming sessions with their psychological advisors. There was no significant difference in computer experience (p>0.806). Table 15: Questionnaire item 5: Experience with computer or console games 5. How many hours per week do you play computer or console games? SotA Proj. Gamif. ( ̅ݔSD) 1.6 (0.9) 1.3 (0.7) 1.8 (1.0) 1: not at all, 2: < 2 hours, 3: 2-5 hours, 4: 5-10 hours, 5: > 10 hours Most participants rarely play at the computer or the console. If they play, they mostly do so during educational gaming sessions with their psychological advisors. There was no significant difference in gaming experience (p>0.206). The five pre-experiment questions show that with little variance there are no signifi cant differences between the three test populations. This is important for the validity of the scenario-specific questions, because major differences would have implied un randomized test populations. 139 Studies and Evaluation: Questionnaire Results and Discussion 7.6.2 Generic Post-Experiment Results While we looked at variances concerning the whole test population in the pre-experi ment questions, for the post-experiment questions the difference between the state-ofthe-Art (SotA) scenario and one of the two augmented scenarios is relevant – i.e. the difference between SotA and Projection or between SotA and Gamification. For this reason two p-values will be discussed after each results table. In several cases the analysis of two questions will be combined: either because they deliberately are contradictory in order to detect the tendency to affirmation (see chap. 7.2.1 Question naire) or because they are thematically related. Table 16: Questionnaire items 6 and 10: Difficulty 6. The help device was easy. SotA Proj. Gamif. ( ̅ݔSD) 1.9 (0.4) 2.4 (0.7) 1.9 (0.3) 4.3 (0.9) 4.0 (0.5) 4.2 (0.5) 10. The help device was unnecessary complicated. ( ̅ݔSD) 1: exactly right, 2: right, 3: neutral, 4: not right, 5: not right at all At first SotA and Projection are focused. Initial results comparing these scenarios have already been discussed in previous work (Korn et al., 2013). The manifest h1-hypothesis is that projection will make the instructions more accessi ble and thus will be perceived as easier, while the h0-hypothesis assumes that there is no significant difference. A t-test shows that h0 is maintained (p>0.027 two-sided). Also h1 has to be revised: although the difference is not significant there is evidence that CAAS augmented by projection are perceived as “less easy” than both SotA and Gamification (∆= ̅ݔ0.5). The antithetic assertion is consistently rejected. Again there is no significant difference (p>0.282 two sided) and the delta to SotA is reduced (∆= ̅ݔ0.2). However, this relativization mainly seems to be due to the impaired per sons’ reluctance to “speak badly” about things (see chap. 7.7 Qualitative Findings). With the gamification scenario the result is clear: the h1-hypothesis that gamification will make the device easier is rejected (p=0.500 one-sided in item 6 and p>0.334 onesided in item 10), so the h0-hypothesis stays intact: there is no significant difference in difficulty. However, this is an important result: since gamification potentially in creases the cognitive workload it is an unexpected finding that CAAS augmented by gamification are not perceived as more difficult than the SotA systems. 140 Studies and Evaluation: Questionnaire Results and Discussion Table 17: Questionnaire items 7 and 8: Acceptance 7. I would use the help device again. SotA Proj. Gamif. ( ̅ݔSD) 1.9 (0.7) 2.0 (0.6) 2.3 (0.7) 1.8 (0.5) 2.4 (0.7) 2.0 (0.5) 8. I would recommend the help device to my colleagues. ( ̅ݔSD) 1: exactly right, 2: right, 3: neutral, 4: not right, 5: not right at all These questions verify if the augmentation is accepted among the impaired users. Asking if a system is recommended to colleagues is a good indirect way of checking the validity of the answer to the direct question because usually a good solution would be recommendable. The results show that the SotA system is preferred. Since it diverges least from the regular workplace this indicates that impaired persons value constancy. However, both augmentations were rated only slightly below, so the h1-hypothesis that projec tion increases acceptance can be denied (p>0.322 one-sided) and the h0-hypothesis that projection has no significant influence on acceptance is maintained. Concerning the high level of divergence with respect to the SotA system this is a positive finding: innovation in the workplace is accepted if the benefits are obvious. Gamification does not increase acceptance, but neither can h0 (there is no significant effect) be reconstituted – indeed there is a marginally significant finding that gamifi cation reduces acceptance (p<0.092 two-sided). In the recommendation question there is a stronger tendency not to acclaim the pro jection system (p<0.107 two-sided) whereas the rejection of gamification is much weaker (p>0.364 two-sided), so the weak tendency against the gamification augmen tation is relativized. Table 18: Questionnaire items 9 and 15: Usability: Ease of Handling 9. I need somebody’s help to use the help device. SotA Proj. Gamif. ( ̅ݔSD) 3.6 (1.1) 3.6 (0.8) 3.7 (0.9) 2.0 (0.5) 2.2 (0.4) 2.1 (0.6) 15. I could use the help device proficiently. ( ̅ݔSD) 1: exactly right, 2: right, 3: neutral, 4: not right, 5: not right at all 141 Studies and Evaluation: Questionnaire Results and Discussion CAAS are meant to improve the work experience of their users and reduce the work load of supervisors. Thus their usability and especially their perceived ease of han dling is essential for their success. Again two incompatible questions were used to reveal affirmative tendencies. Most participants clearly felt that they would not need help to use the assistive system. The differences regarding this issue are not significant at all (p>0.871 two-sided in the comparison between SotA and Projection and p>0.877 two-sided in the compari son between SotA and Gamification). As the variance shows, rejecting help (item 9) was an additional psychological barrier for some of the impaired while expressing competency was not. So with very little variance the proficient use was confirmed by all participants. Differences between the scenarios are minimal and not significant (p>0.260 two-sided in the comparison between SotA and Projection, p>0.537 twosided in the comparison between SotA and Gamification). Table 19: Questionnaire items 13 and 16: Usability: Learning 13. Working with the help device can be learned quickly. SotA Proj. Gamif. ( ̅ݔSD) 1.8 (0.6) 2.2 (0.6) 2.4 (0.6) 3.6 (0.9) 3.6 (0.9) 3.4 (0.8) 16. I had to learn a lot before I could work with the help device. ( ̅ݔSD) 1: exactly right, 2: right, 3: neutral, 4: not right, 5: not right at all We already pointed out the importance of a perceived ease of use in the items regard ing ease of handling. Since this subject is central for the acceptance of CAAS by the users and their supervisors, two additional incompatible questions directly focus on learning. Similar to the questions focusing on acceptance (see Table 17: Questionnaire items 7 and 8: Acceptance) the impaired users are aware of the status quo and believe that the least divergent system can be learned most quickly. Accordingly the mean results of item 7 (“I would use the help device again”) and item 13 are very close in all three scenarios. Both the CAAS augmented by projection (p<0.032 two-sided) and the system aug mented by gamification (p<0.001 two-sided) are rated significantly lower when com pared to the SotA system. So in this case the h0-hypothesis claiming that there is no effect by augmented CAAS on learning is falsified. 142 Studies and Evaluation: Questionnaire Results and Discussion The results on the contradictory questions differ. While most users confirm their basic notion and tend to answer that they had not “to learn a lot” before they could use the CAAS, there is no significant difference between the grade of denial. The difference between SotA and Projection is insignificant (p>0.504 two-sided) and so is the differ ence between SotA and Gamification (p>0.595 two-sided). The almost identical values indicate that this question was answered automatically – the “confession” that the own mental capacities were at their limits (“I had to learn a lot…”) in this item seems much harder for the impaired persons than the evaluation of the positive framing of the question. Table 20: Questionnaire items 11 and 12: Generic User Interface 11. The indicator for picks was easy. SotA Proj. Gamif. ( ̅ݔSD) 1.9 (0.7) 1.8 (0.4) 2.0 (0.0) 3.5 (1.2) 3.2 (1.0) 3.4 (0.9) 12. I found it tiring to confirm each step with a button. ( ̅ݔSD) 1: exactly right, 2: right, 3: neutral, 4: not right, 5: not right at all As described in the implementation (see chap. 6.4.1 Visual Output and Projection) two elements were projected in all three scenarios: the “next” and “previous” naviga tion and the indicators for picks. While there is no technical difference in how these “generic user interfaces” were projected and interacted with, there might be an influ ence from the other interfaces – e.g. in Projection additional elements are projected into the workspace. However, in spite of the additional projections in the work area, there is no significant difference in the perception of the indicator between SotA and Projection (p>0.432 two-sided) and between SotA and Gamification (p>0.540 two-sided), so the pick in dicator was considered to be “easy” by the complete test population and does not have to be considered in the further analysis. A look at the second permanently projected element – the step navigation – shows a similar picture: neither is there a significant difference in the perception of the navi gation between SotA and Projection (p>0.477 two-sided) nor is there a significant difference between SotA and Gamification (p>0.886 two-sided). The tendency to “neutral”, in this case a mild rejection of the element, shows that the need to confirm single work steps was indeed a minor disturbance – but since the whole test population faced it, it will not be considered. We can conclude that both generic user interface elements did not have a significant impact on the perception of CAAS. 143 Studies and Evaluation: Questionnaire Results and Discussion Table 21: Questionnaire item 14: Self-Perception and Strain 14. I found the work with the help device exhausting. SotA Proj. Gamif. ( ̅ݔSD) 4.0 (0.7) 4.0 (0.6) 3.8 (0.8) 1: exactly right, 2: right, 3: neutral, 4: not right, 5: not right at all This important question was to check if the augmented systems increase the perceived strain. The almost identical values in SotA and Projection show that this does not apply in their case (p>0.814 two-sided). The mean strain in the gamification scenario is a bit lower – nevertheless the differ ence is not great enough to be statistically significant. The h1-hypothesis that gamifi cation decreases strain can be denied (p>0.201 one-sided), so the h0-hypothesis that gamification has no significant influence on strain is maintained. 144 Studies and Evaluation: Questionnaire Results and Discussion 7.6.3 Specific Post-Experiment Results While the 11 items discussed in the last chapter were completely identical, the follow ing questions differed between the scenarios. Still the first three questions only differ with respect to the method used to display the instructions (monitor versus projection) and thus are comparable, whereas the subsequent questions (three in the projection scenario and five in the gamification scenario) were completely scenario-specific. Table 22: Questionnaire items 17, 18 and 19: Instructions 17. I found the instruction on the monitor [Projection: left of me on the table] complicated. SotA Proj. Gamif. ( ̅ݔSD) 4.0 (1.0) 3.8 (0.7) 3.8 (0.6) 1.8 (0.4) 2.2 (0.4) 2.1 (0.4) 3.9 (0.9) 2.0 (0.3) 3.5 (0.8) 18. I found the instruction on the monitor [Projection: left of me on the table] easy. ( ̅ݔSD) 19. I found it tiring to always check the instruction on the monitor above me. [Projection: I liked having the instruction directly to my left on the table.] ( ̅ݔSD) 1: exactly right, 2: right, 3: neutral, 4: not right, 5: not right at all Since both the instructions and the task were identical in all three scenarios, differ ences in the perception of the instructions are largely attributable to the method of display. However, in Projection an additional 1:1 model was projected into the center of the workspace (see below), so the instructions in this scenario were redundant. Regarding the perception of the instructions’ complexity there are no significant dif ferences between the scenarios (SotA/Projection: p>0.471 two-sided; SotA/Gamifi cation: p>0.571 two-sided). The control question about the instructions being easy (item 18) shows that SotA was considered to be significantly easier than Projection (p<0.0072 two-sided) although there is no significant difference between SotA and Gamification (p>0.033). At first these results are surprising. However, we already found that negative utterances are avoided by the test population (see Table 16: Questionnaire items 6 and 10: Difficulty: “The help device was easy” and “The help device was unnecessary complicated”). This explains why there are no significant differences in item 17. When feeling “on 145 Studies and Evaluation: Questionnaire Results and Discussion the safe side” as in item 18, the answers show that the display of instructions in Pro jection is considered “less easy”. Considering there is strong evidence that the subjects value constancy (see Table 17: Questionnaire items 7 and 8: Acceptance, Table 19: Questionnaire items 13 and 16: Usability: Learning), the new way of displaying the instructions probably made them “less easy” or better: harder to learn. However, the time measurements and the subsequent questions show that most subjects were well capable of adapting to projected instructions. In item 19 the advantages of projection are fully unfold. We supposed that frequently having to look up to the monitor with the instructions and losing the cognitive focus (SotA and Gamification) would bother the workers. However, they seem to be used to this annoyance from paper-based instructions and tend to “not find it tiring”. As can be expected with largely identical systems the differences between SotA and Gamification are not significant (p>0.201 two-sided), although there is a tendency to perceive gamification as more tiring (∆ = ̅ݔ0.4). This is attributable to the animated gamification component which demands additional attention and cognitive focus each time the participant looks at the instructions. With very little variation the participants agreed that they liked having the instructions projected directly on the table. At first this seems contradictory to the perception of SotA as significantly easier than Projection (see above: item 18). However, it has been shown that the participants value constancy but acknowledge improvements. So while they had to learn more, they almost unanimously confirmed the advantages of projec tion they experienced during the assembly phase. Table 23: Questionnaire item 20 (Projection): Instruction 20. The instruction left of me on the table disturbed my work. Proj. ( ̅ݔSD) 4.1 (0.8) 1: exactly right, 2: right, 3: neutral, 4: not right, 5: not right at all It was an important aim of the survey to be sure about the perceived benefits of pro jection. For this reason a question contradictory to item 19 was integrated (“I liked having the instruction directly to my left on the table”). The workers strongly rejected this assertion: in the complete survey next to one occurrence of 4.3 and two occur rences of 4.2 the value 4.1 (three occurrences) marks the strongest negation. This result supports the finding from item 19: projected instructions are considered very helpful. 146 Studies and Evaluation: Questionnaire Results and Discussion Table 24: Questionnaire items 21 and 22 (Projection): Additional Product Model 21. The model in the center of the workspace helped me. Proj. ( ̅ݔSD) 2.0 (0.2) 22. The model in the center of the workspace bothered me. ( ̅ݔSD) 4.1 (0.4) 1: exactly right, 2: right, 3: neutral, 4: not right, 5: not right at all With the second lowest variance in the survey, the subjects agree that it was helpful when an additional 1:1 model of the current state of the product component is pro jected into the center of the workspace (see chap. 7.2.2 Assembly Task). When asked the opposing question they responded accordingly and rejected the asser tion. These findings strongly support the finding of the previous projection-specific items 17 to 20 that projection – in spite of the workers’ belief that the system is more difficult to learn – is highly beneficial for impaired workers. Table 25: Questionnaire items 20, 21 and 24 (Gamification): Game 20. I liked the game on the monitor. Gamif. ( ̅ݔSD) 1.9 (0.4) 21. The game on the monitor bothered me. ( ̅ݔSD) 4.2 (0.5) 24. I would like to always have the game at work. ( ̅ݔSD) 2.4 (0.9) 1: exactly right, 2: right, 3: neutral, 4: not right, 5: not right at all As in several other cases two oppositional items are used. The results are consistent: while most participants agree that they liked the game, they also agree that the game did not bother them. An approval rate of 1.9 marks the second-highest agreement within the post-experiment part of the questionnaire (there are three occurrences of 1.8 and four occurrences of 1.9). Thus the approval level of Gamification is similar to the approval of Projection. 147 Studies and Evaluation: Questionnaire Results and Discussion In the third question (item 24) we wanted the test subjects to step back from the expe rience in the experiment and imagine a workplace permanently augmented by gami fication. The variance is greater with that question – and although the result is still positive it is significantly different from the acceptance of the game itself in item 20 (p<0.041 two-sided). Here the attitude to look favorably at innovation while retaining a mildly skeptical attitude again becomes obvious. Table 26: Questionnaire items 22 and 23 (Gamification): Comments 22. The comments on my work motivated me. Gamif. ( ̅ݔSD) 1.9 (0.4) 23. The comments on my work disturbed me. ( ̅ݔSD) 4.1 (0.7) 1: exactly right, 2: right, 3: neutral, 4: not right, 5: not right at all Comments combine two elements (see chap. 6.5 Gamification Component): a visual element starting with the dissolving of the built-up brick rows, followed by a smiley indicating the performance after each sequence audio feedback which is given briefly after each process and in an extended form after the completion of a sequence The acceptance of the comments resembles the acceptance of the game itself: the same high approval rate with the same low variance. Also the negative framing of the ques tion is consistently negated. Thus the approval rate of comments in Gamification is similar to the approval of Projection. 148 Studies and Evaluation: Qualitative Findings 7.7 Qualitative Findings In an experiment with 60 impaired persons, behavior can be measured extensively – but still some experiences are hard to describe in numbers. A first qualitative finding was that all participants were very co-operative. Not a single participant asked for money – although they all gladly accepted food and sweets prepared on a trolley table at the room’s entrance. While some were afraid of the new situation at the beginning and others were shy in the personal communication, everybody who was able to use the system left the room in a positive mood – and this applies to all three test scenarios. This positive disposition may partly be attributable to the fact that “they made it” – but for the system designers it was important to see that a novel approach is not flatly rejected or results in a negative mood swing. As described in the design of the questionnaire (see chap. 7.2.1) we deliberately added contradictory assertions (e.g. item 6: “The help device was easy” and item 10: “The help device was unnecessary complicated”) to detect affirmative tendencies. We found that there was no clear affirmative tendency – however we experienced a “po lite” tendency, i.e. the impaired persons’ reluctance to “speak badly” about things: while advantages are clearly pointed out, the elevation of disadvantages is avoided. This becomes very clear with questions regarding learning (Table 19: Questionnaire items 13 and 16: Usability: Learning): the positive framing of the question (…can be learned quickly) results in significant and highly significant findings whereas the neg ative framing (“I had to learn a lot…”) leads to an almost identical pattern of rejection. In the discussion with the participants they also were reluctant to talk about their own shortcomings. Although the participants are clearly impaired, in this respect they react just like everybody else: they want to talk about good things and success rather than problems and failures. Hierarchy and status are important: Manufacturing parts for Audi is perceived as “better” than manufacturing parts for an unknown company. Of ten a proud impaired worker would say something like “Me – Audi” meaning that he had a top job within the sheltered work organization. The projection was very well appreciated by the subjects. It seemed to be natural, whereas having the instructions on a monitor looked antiquated in direct comparison. Also the problem of shadows was smaller than we anticipated since the participants soon learned to avoid them. The small model in the center of the workspace was an essential improvement. Several participants refrained completely from looking at the full instructions projected to their left and instead simply focused on mirroring the 1:1 model. Studies and Evaluation: Qualitative Findings 149 The participants seemed to react more to the gamification component’s auditory feed back than to the visual feedback: the Tetris-like production game on the monitor was rarely looked at during the work processes. However, the evaluation at the end of each sequence with the smiley and the longer spoken feedback was explicitly looked at and commented on. The smileys after each sequence obviously were noticed because the cognitive focus already shifted from the micro to the macro level when the assembled product was placed behind the boxes. Figure 77: User checking the instruction and the visual gamification element. Still the participants really seemed to value and enjoy the feedback. Many started talking back, especially when the feedback implied room for improvement, e.g. the feedback “you did that better before” frequently resulted in denials. Obviously the idea of developing CAAS as companions would be highly appreciated by the impaired users. While the current gamification setup drew the focus away from the workspace (see Figure 77) future approaches need to bring the visual gamification elements closer to the workspace as the center of the user’s attention. The combination of projection and gamification can easily achieve that (see chap. 8.3.1 Technical Perspectives). 150 Studies and Evaluation: Discussion 7.8 Discussion In this discussion we resume the most important quantitative and qualitative findings and analyze how they are related. 7.8.1 Similarities between the Scenarios When looking at the survey results at a meta-level, it is striking how many central HCI issues in the three scenarios were seen as equivalent by the impaired production workers: there are no significant differences between the SotA and the augmented scenarios in overall acceptance or in handling. Most participants state that they can use these systems proficiently and without additional help. This is supported by the qualitative findings: the impaired persons accepted the systems as they were. With respect to usability there is a clear finding that neither Projection nor Gamification are perceived as significantly more difficult. One could argue that acceptance is a function of the users’ speed and error rate, so low error rates (e) and low sequence times (t) imply a high acceptance. Indeed there is a correlation, but it is not strong with SotA rt=0.18 and re=0.20, Gamification rt=0.26 and re=0.13 and Projection rt=0.03 and re=0.35 (example values for item 6: “The help device was easy”). However, it is interesting that the strongest correlation is linked to errors in Projection (see below). The hypothesis that gamification decreases strain could not be statistically confirmed. Although there is a recognizable tendency in that direction and the gamification sce nario is rated the least exhausting, the differences are not accentuated enough to be significant. Nevertheless both the adapted feedback and the production game are rated with the second-highest agreement level and show the lowest standard deviation within the post-experiment questions (= ̅ݔ1.9, SD=0.4). On the qualitative level the adapted visual and auditory feedback resulted in the CAAS being perceived as a com panion – a pleasant finding that shows how CAAS can be integrated to also generate ethical benefits. When comparing the individual users, there was no significant effect of the augmen tations on production time. While there are clear trends towards shorter sequence du rations, the high variance between the users makes the effects of augmentation by projection (mean time reduction of 7.8%, p>0.259 one-sided) and gamification (mean time reduction of 12.5%, p>0.110 one-sided) statistically insignificant. With regard to the error rate CAAS augmented by projection which do not provide feedback on product-specific errors increase the error rate by an absolute 6.1% and a relative 28.7% which again is not statistically significant (p>0.186 one-sided). Studies and Evaluation: Discussion 151 To resume, the overall findings are positive: the impaired production workers are open for innovation and look at new approaches without prejudices – and obviously the augmentations in the CAAS prototype worked well enough to be acceptable. In fact especially projection already seemed so well integrated and “natural” that having the instructions on a monitor looked antiquated in direct comparison. 7.8.2 Differences between the Scenarios Generally the augmentations’ comparatively low quantitative impact on speed and partly also on errors was unexpected. Methodically this is due to the high variance between individual impaired users. To balance this effect, we also looked at how the groups’ means developed and checked if there are deviating sub-groups. When it comes to error rates the results are more pointed: CAAS not providing prod uct-specific feedback augmented by gamification increase the error rate by an absolute 10.5% and relative 46.5% which is significant in spite of the high user variance (p<0.052 one-sided). All the results can be acuminated by focusing on the temporal development of mean production speed and mean error rate in the eight sequences instead of the individual users. While even in this aggregated perspective CAAS with projection do not signif icantly reduce the mean sequence completion times (p>0.200 one-sided), CAAS aug mented by gamification do so (significantly with p<0.064 one-sided). When analyzing the development of mean sequence error rates (see Figure 72) there clearly are signif icantly more errors with both Projection (p<0.003 one-sided) and Gamification (p<0.00001 one-sided). The Projection data demanded a more thorough analysis. While projected instructions were generally looked at favorably, they were also perceived as significantly “less easy” than instructions on the monitor. While the additional 1:1 model was a highly appreciated augmentation of the production process, this deviation from the known standard noticeably increased the perceived overall complexity. This finding contradicts the intuitive assumption that the projection of information directly into the workspace will be perceived as an improvement. However, the high correlation between error rate and perceived system complexity mentioned above shows that projection had a strong impact on the population. And indeed we found that projection has a catalytic effect: it significantly increases the speed of the faster workers (p<0.061 one-sided) and even significantly reduces their error rate (by 45.8% compared to the faster SotA workers, p<0.053 one-sided). At the same time the slower workers perform worse with CAAS augmented by projection than with the SotA sys tem. This aptitude for the successful use of CAAS with projection is not just a function 152 Studies and Evaluation: Discussion of the users’ wage level, the performance indicator index used by sheltered work or ganizations (p>0.147 one-sided) – it relates to a different feature. Compared to the SotA system both augmented CAAS were seen as significantly more difficult to learn. So while the augmented systems are perceived as easy to use once they are mastered (as there are no significant differences in overall acceptance), the perceived learning curve is steeper with Projection and even steeper with Gamifica tion, where the cognitive focus is frequently drawn to the gamification element. How ever, we observed a clear trend towards making less errors in the faster subgroup (Figure 76), so augmenting workplaces with projection also seems to trigger learning. To resume, CAAS are well-accepted and appreciated by the users. Projection makes monitors look old-fashioned and the gamification of work is both appreciated and enjoyed. While these augmentations appear to speed up users (which can be shown in the aggregated view on the development of sequence time) in their current form they also lead to increased error rates. The only exception are faster users of the projection augmentation who perform significantly better in all areas – probably because they are also faster learners. However, for CAAS to be successful in sheltered work organizations and in produc tion companies they will need to be able to detect product-related errors. Currently the augmentations draw cognitive processing power away from the assembly so that the speed gains can be attributed to a speed-accuracy trade-off. This effect has to be counter-balanced by a real-time analysis of the product state (see chap. 8.3.1 Tech nical Perspectives). 153 Conclusion: Summary of Research Contributions 8 CONCLUSION Context-aware assistive systems (CAAS) will change the way we work. Like route guidance systems have changed the way we drive in unknown areas (and the amount of time we spend in preparation), CAAS will permanently change work. In the case of work in manual production, errors will be addressed “in the making” and persons with cognitive or motoric impairments will be able to remain in active production jobs longer. At the same time CAAS allow to detect changes in performance and adjust the level of challenge and the feedback. Ideally this results in a worker who remains in the flow channel, i.e. an area where skill and challenge converge and a high level of concentration is accompanied by a feeling of satisfaction and accomplishment (see chap. 2.4.3 Flow). However, research on CAAS in production environments is just beginning. Neverthe less the pioneering work presented here sheds light on several important aspects which have been introduced in the research questions (see chap. 1.3). 8.1 Summary of Research Contributions In the following important results regarding the research questions (RQ) are summa rized. For a more detailed discussion the relevant chapters are indicated. Table 27: RQ 1: Requirements of CAAS. No. Research Question (RQ) Chapter RQ1 Which requirements should CAAS in production environments address? 4 Although common standards provide a proven and tested framework for HCI, the use in production with impaired persons results in several constraints and adaptations re garding the focus of interaction, the demand for total quality and the ideal of universal design. From the starting requirement of implicit interaction the requirements of automatic detection of movements as well as the detection of the current work state and speed were derived. The easy-to-use motion recognition systems used by CAAS address these requirements and allow an unprecedented real-time analysis of work actions. While the requirement to move work-relevant information closer to the worker could be addressed by applying state-of-the art projection technology, the requirement to 154 Conclusion: Summary of Research Contributions detect errors (resulting from the demand for total quality in production) could not be addressed in the described CAAS prototype – it presupposes either finger or object recognition and remains an area for future research. The requirement of adaptation to the user’s competence level results from general HCI standards and the paradigms of universal design and universal access. While the ad aptation can be realized with the motion data alone, it would benefit from additional sensors recognizing excitement (e.g. facial emotion detection, heart rate detection). The motion data also allows new approaches like the integration of motivating ele ments into the comparatively monotonous work in manual production. Gamification can be used to keep the user’s awareness up and thus to prevent errors. Furthermore motivation and fun during work are ethically desirable. While this requirement could be addressed, it is strongly linked to adaptation. Thus it also would strongly benefit from additional sensors recognizing emotions. The final requirement is an ethical necessity: the protection of the user’s personal data. Since CAAS “come very close” to the user, analyzing movements and potentially even emotions, these data have to remain strictly separated from the regular workrelated data. Meeting this requirement is a future task during the large-scale integra tion of CAAS in industry contexts. Table 28: Generic model for CAAS. No. Research Question (RQ) Chapter RQ2 How can the requirements be prescinded to a generic model for CAAS? 5 The CAAS model mainly draws from three concepts: the established HAAT-model for assistive technology, the framework for an adaptive game system and the flow concept. The HAAT model is adapted so that the human-technology interface can be realized by projection, the environmental interface can be realized by natural interac tion (NI) based on motion recognition and the activity output can be enriched by gam ification. The adaptive game system is re-interpreted so that NI can be used as an instrument that allows game-like interaction in non-gaming contexts. The concept of flow is expanded by “flow curves” which imply that an activity has to be designed in phases to be permanently motivating. The model aims to show the parallels in processing information. Both the human and the CAAS share an environmental interface consisting of sensors on the input side and various actors on the output side. The CAAS input receives data which is prepared 155 Conclusion: Summary of Research Contributions for the interpreter to create a fitting model of the current state of the user, e.g. a de tected sequence of errors results in a lower mental score, a detected stress symptom shifts the flow curve towards arousal. The modeled user state eventually results in an adjustment of the operational mode, e.g. the speed of production. The behavior after an adaptation will indicate if the human state was modeled cor rectly, so the iterative interpretation of behavior changes can be used to correct errors in modeling. The CAAS generator adapts the interventions: the gamification compo nent, the instructions and the feedback. The adapted interventions are distributed over various output channels. Table 29: Implementation of motion recognition No. Research Question (RQ) Chapter RQ3 How can motion recognition be implemented in CAAS? 6.3, 6.4 Motion recognition allows implicit interaction. An ideal CAAS as described in the model recognizes intricate finger movements. However, at the time of the implemen tation none of the low-priced motion recognition systems did provide the spatial and temporal resolution required for this level of detail. Thus errors related to picking components could be detected, while product-related errors like a side-inverted as sembly remained undetectable. A further technical restriction was the impossibility to use skeletal joint detection because the corresponding middleware requires the sensors to scan horizontal rather than vertical areas. Instead of using the joints, the CAAS prototype analyzes interconnected areas by us ing the changes of z-values (adjoined-spheres approach). These 3D trigger areas can be referenced and thus the system’s sensitivity can be adjusted to detect movements in areas already partially filled. This approach is generic and can be implemented in other scenarios with similar research or implementation interests. The approach was implemented in the software “Assistive Systems Experiment De signer”. In its designer component the steps of a production sequence can be mapped by assigning causes and effects. The corresponding runner component monitors the trigger areas during work and creates a video of the workspace for documentation purposes. Whenever an interactive area is triggered a log entry is generated. Like the adjoined-spheres approach the system is generic and allows to use motion recognition to analyze any kind of interaction or work activity in confined spaces. 156 Conclusion: Summary of Research Contributions Table 30: Implementation of projection. No. Research Question (RQ) Chapter RQ4 How can projection be implemented in CAAS? 6.4.1, 7.2.2 Projection allows to move work-relevant information from the user’s periphery closer to the center of activity. This could be realized by applying state-of-the art technology: the visual elements (instructions, controls for navigation and indicators for picks) were projected directly into the workspace. A challenge was that the prototype CAAS had to be able to administer three visual output devices simultaneously, so hardware issues had to be addressed. Furthermore the screen coordinates of all graphic devices had to be synchronized. Finally the pro jector’s tilt required by the physical setup implied a keystone distortion which had to be corrected by a non-affine transformation in real-time. Figure 78: User comparing the assembled product with the 1:1 model. On the practical level a simple idea proved very successful: projecting a 1:1 model of the correct result of the assembly with the current step marked by a green arrow into the center of the workspace (see Figure 78). Several users relied completely on that simplification – and we got the impression that the excellent results of a part of the SotA population have been furthered strongly by this visual aid. 157 Conclusion: Summary of Research Contributions Table 31: Implementation of gamification. No. Research Question (RQ) Chapter RQ5 How can gamification be implemented in CAAS? 6.5 Gamification allows to keep a user’s awareness up during repetitive and monotonous work – and this increased awareness helps to prevent errors; also increasing motiva tion and fun during work is ethically desirable. The basis of the gamification implementation is the motion data that allows measuring the assembly times. Additional sensors allowing facial emotion detection or heart rate detection were not implemented and are a subject of future research. The basic design approach was to address the conditions for flow. This was achieved by creating a “production game” where each work process is visually represented as a brick. During the corresponding work process the active brick moves downwards while its color changes from green to red. This time assessment is based on the user’s personal mean of a limited number of previous assemblies of the identical product. This hypothetical production speed is also visualized by a “shadow brick” moving down at exactly this speed. If a product is completed, all bricks disintegrate. After each process and after each completed product the user receives multimodal feedback (visual and audio). The multimodal feedback already gives the user the essential feeling of self-efficacy. However, in long-term operation gamification can also be used to actively influence and support the flow curve. The assembly times of the recent sequences then are used by the interpreter to determine the user’s current position on the flow curve and detect trends and phase changes, e.g. from flow to arousal. While such changes are perfectly normal, an intervention can be created if the phase of arousal or control last too long or the trend indicates that the user leaves the flow channel (e.g. when moving from arousal to anxiety). This can be achieved by adapting the production cycle and the speed of the shadow brick. As an example an excited user can be calmed by slowing the shadow bricks. This leads to more “successful” assembly processes unmistakably mirrored by green bricks and positive feedback. Thus a user’s self-confidence and motivation can be actively supported. 158 Conclusion: Summary of Research Contributions Table 32: Quantitative impact of augmentations on speed and quality. No. Research Question (RQ) Chapter RQ6 What is the quantitative impact on work speed and quality when work is augmented by projection and gamification? 7.5 A main issue for the analysis of effects when working with impaired persons is the high variance between the users. If all individual user times are taken into account neither projection (mean time reduction of 7.8%) nor gamification (mean time reduc tion of 12.5%) have a statistically significant impact on time. However, when looking at the data at an aggregated perspective and analyzing how the groups’ mean speed developed over the eight production sequences, there are sig nificant differences: while CAAS with projection cannot be shown to reduce the mean sequence times (p>0.200 one-sided), CAAS using gamification significantly reduce the production speed (p<0.064 one-sided). With the error rates the effects are gener ally more accentuated: CAAS augmented by projection not providing feedback on product-specific errors increase the error rate by an absolute 6.1% and a relative 28.7%. If all individual users are taken into account, this is not enough to be statisti cally significant (p>0.186 one-sided). CAAS augmented by gamification not provid ing product-specific feedback increase the error rate by an absolute 10.5% and a relative 46.5% which is statistically significant in spite of the high user variance (p<0.052 one-sided). When looking at the data at the aggregated perspective to analyze how the groups’ mean error rate developed over the eight production sequences, both augmented sys tems resulted in significantly higher error rates (p<0.003 one-sided with projection and p<0.00001 one-sided with gamification). In summary the currently observed effects of gamification and projection on work speed and quality are negative because the increased speed is compensated by an in creased error rate so the total quality paradigm is violated. Obviously the users’ ben efits from projection and the motivational gain from gamification were transformed into speed, and since no feedback on quality was given a speed-accuracy trade-off occurred. Potentially the positive effects on speed can be preserved if future CAAS manage to create interventions when product-specific errors happen (see chap. 8.3 Future Research). When looking at sub-groups, there is an important finding regarding the augmentation by projection. The faster part of this group differs from all other sub-groups: the mean error rate is only 14.7% – given that the complete population’s mean error rate is 29.1% these faster users made only about a third of the mistakes. Compared to the 159 Conclusion: Summary of Research Contributions faster users in the state-of-the-art population this is a relative error reduction of 45.8%. The hypothesis that projection decreases the error rate of above average workers can be accepted (p<0.053 one-sided). This is especially important because these workers did not receive feedback on product-specific errors. Probably they made good use of the 1:1 model projected in the center of the workspace. Finally there is a clear trend towards making less errors in this group (see Figure 76) so augmenting a workplace with projection also seems to trigger learning and thus continually improves quality. Table 33: Quantitative impact on users: No. Research Question (RQ) Chapter RQ7 What is the quantitative impact on users when work is augmented by projection and gamification? 7.5 When looking at the results of the survey at a meta-level it is astounding how many important HCI issues in the different scenarios were seen as equivalent by the im paired production workers: there are no significant differences in overall acceptance, in usability and in handling. Most participants feel that they can use these systems proficiently and without additional help. These are positive findings: the fact that there are no significant differences in these decisive aspects shows that the production workers are open for innovation and look at new approaches without prejudices. Both the feedback and the production game are rated with the second-highest agree ment level and the lowest standard deviation within the post-experiment question (= ̅ݔ1.9, SD=0.4). However, the hypothesis that gamification decreases strain could not be confirmed. Although there is a recognizable tendency in that direction and the gamification scenario is rated the least exhausting, the differences are not significant. While Projection was generally looked at favorably, projected instructions were con sistently perceived as significantly “less easy” than the instructions on the monitor. While the additional 1:1 model was an augmentation highly appreciated by all users, it was also recognized as a deviation from the known standard that increased com plexity. This is interesting in the light of the catalytic effect projection had on the performance: while some workers learned to use the additional aid and improved sig nificantly, others were confused and performed worse than with the SotA-system. So while CAAS are perceived as easy to use once they are mastered, both augmented systems were perceived as significantly more difficult to learn. The ability to learn or simply adapt to a new system obviously is crucial for the success. 160 Conclusion: Summary of Research Contributions Table 34: Qualitative impact on users: No. Research Question (RQ) Chapter RQ8 What is the qualitative impact on users when work is augmented by projection and gamification? 7.7 The survey’s positive evaluation of the CAAS acceptance, usability and handling was confirmed by our qualitative findings: everybody who was able to use the system left the room in a positive mood. This positive disposition may partly be attributable to the fact that “they made it” – but for the system designers it was important to see that a novel approach is not flatly rejected or results in a negative mood swing. The projection was very well appreciated by the subjects. It appeared natural, whereas looking at the instructions on a monitor seemed antiquated in direct comparison. The small model in the center of the workspace was an essential improvement. Several participants refrained completely from looking at the full instructions projected to their left and focused completely on this 1:1 model. The participants seemed to react more to the gamification component’s auditory feed back than to the visual: the production game on the monitor was rarely looked at dur ing the work processes. However, the participants valued and enjoyed the feedback: the evaluation at the end of each sequence with the smiley and the longer spoken feedback were explicitly looked at and commented on. The adapted visual and audi tory feedback resulted in the CAAS being perceived as a companion – a pleasant find ing that shows how CAAS can be integrated to generate ethical benefits. The current gamification setup drew the user’s focus away from the workspace; future approaches will need to integrate such elements into the workspace closer to the center of the user’s attention. This can be achieved by combining projection and gamifica tion. Table 35: Ethical dimension of CAAS: No. Research Question (RQ) Chapter RQ9 What is the ethical dimension of CAAS and how can they be designed to be humane? 2.3, 8.2 The ethical dimension of CAAS is discussed in the following sub-chapter. Conclusion: Ethical Implications: Towards Humane Work 161 8.2 Ethical Implications: Towards Humane Work The introduction stated that advances in technology must not increase the gap between people but can open radically new perspectives and even lead to the “full inclusion of [these] individuals […] in the mainstream of society” (Cook, 2010). While new tech nologies are readily accepted by many users, critics have put forward the argument that the recent advances in AI and automated systems will in due time allow robots (or autonomous machines) to take over all forms of routine human work – and that this development would finally result in an unemployment crisis (Ford, 2013). This future scenario describes the ancient human fear that one day the potentials of tech nology will be used to make humans obsolete. While this potential future scenario cannot be denied, there is an alternative: the use of the mentioned advances in the field of assistive technology to enhance human work to the point where it becomes economically valid, meaningful and enjoyable. Even more importantly, this “positive” scenario is not a distant vision – it can be realized with today’s technology. Context-aware assistance systems (CAAS) with augmenta tions like projection and gamification can help to shape work activities according to the user’s abilities while increasing productivity and preserving the sense of complete ness previous forms of work had (see chap. 2.3.2 Value of Work). They have the potential to empower impaired and elderly workers to do more complex work or re main in active jobs for a longer time thus addressing the demographic change. However, CAAS also come at a cost. In order to adapt to a user and support her or him as good as possible, these systems need to “come close”: they need to know as much about the person as possible. As discussed one of the requirements for robust product-specific error detection is high resolution real-time motion analysis, but even heart rate and emotion detection are technological possibilities being explored right now. Like a good psychiatrist a futures CAAS would be more “aware” of a user’s physical, mental and emotional state than we typically are ourselves. Should continuous data management – the ideal of the virtual factory and of cyber physical systems (see chap. 2.6.3) – really integrate “human resources”? Is it ethical to observe a user’s emotions and create a model of the mindset? Does “full inclusion” justify to use this data to deliberately manipulate a worker with gamification ele ments? Is it ethically tolerable that such gamification elements replace intrinsic re wards with explicit ones? Such questions have already been addressed in the MEESTAR-model (see chap. 3.5 Ethical Standards for Assistive Technology) with respect to the areas care and health. In work contexts assistive systems like the CAAS described in this work have never 162 Conclusion: Ethical Implications: Towards Humane Work been used before – and neither have their ethical implications been discussed or eval uated. The ultimate question is the value we attribute to work: is it an integral part of a good life and is there a “right to work”, as the UN Convention on the Rights of Persons with Disabilities suggests (see chap. 2.3.3)? If we follow this line of argu ment, using all available technological means to support persons to be able to work might become a moral obligation, similar to the medical domain where the preserva tion of life outweighs many other aspects. If our societies take this new “right to work” seriously, humane work in the long run will become not only an ethical recommenda tion but a legal necessity. Once we take the elemental importance of work for granted the next question is how technology like CAAS, which highly impact the way we work, can be integrated in the most beneficial way: how can a balance between the technological possibilities of a quantified self and the desire for freedom, privacy and autonomy be achieved? These questions require much interdisciplinary research and long-term studies. While the integration of CAAS into a company’s product-related data systems is per ceived as essential (see chap. 4.3 Requirements Study) the movement analysis (phys ical state) should be treated differently. This applies even more to the modeled mental and emotional states. Based on our quantitative and qualitative findings, we propose the following guideline of ethical standards for future CAAS: 37 Product-related data generated by the CAAS environmental interface37 dur ing the work process which could also be gained by subsequent camera in spection of products is allowed to be transferred to enterprise data management systems (like ERP or PLM). Data reflecting a person’s physical disposition (e. g. movement, stance) may only be used to improve the product (e.g. detecting errors) or to help the worker (e.g. indicating a malposition). It remains black-boxed within the CAAS and is deleted as soon as possible. Data reflecting a person’s mental or emotional state may only be used to adapt the CAAS operation mode or to generate adapted interventions (gam ification, instructions and feedback) that are eligible for increasing the worker’s motivation and well-being. It remains black-boxed within the CAAS and is deleted as soon as possible. The terminology used is based on the model of CAAS presented in chapter 5. Conclusion: Future Research 163 8.3 Future Research This sub-chapter in based on the following publications: Funk, M., Korn, O., & Schmidt, A. (2014). An Augmented Workplace for Enabling User-Defined Tangibles. In CHI ’14 Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems. New York, NY, USA: ACM. Funk, M., Korn, O., & Schmidt, A. (2014). Assisitive Augmentation at the Manual Assembly Workplace using In-Situ Projection. In CHI ’14 Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems. New York, NY, USA: ACM. CAAS represent a new line of research – especially in the industrial domain. While this work describes first steps into this area, several questions could not be answered with certainty – and additional questions came up. This applies both to the technical side and the ethical implications of CAAS in workplaces. Both areas are portrayed in the following. 8.3.1 Technical Perspectives At the time of the implementation (mid 2012) none of the low-priced motion recog nition systems did provide the spatial and temporal resolution necessary for a robust detection of intricate finger movements as common in manual production. Thus the implementation had to be simplified in that respect and no real-time product-specific error detection with corresponding interventions was integrated (see chap. 6.3.1 Technical Restrictions and Solution). Thus a major technical issue for future research is implementing real-time productspecific error detection. This can be achieved by supplementing motion recognition, which essentially focuses on human body tracking, with object recognition: the result of a work step is compared with a representation of the intended product at that time. Another possibility is the integration of more powerful low-cost motion recognition sensors, e.g. the improved Kinect used in the Xbox One which uses a real time-of flight camera (see chap. 3.1 Motion Recognition). To address these issues, in the fol low-up project motionEAP the 3D-spaces of the Kinect and the Leap Motion sensor are already being combined (Funk et al., 2014a, 2014b) to allow robust error detection for CAAS. 164 Conclusion: Future Research Figure 79: A future setup with multiple sensors combining motion and object recognition. The future implementation of excitement and emotion detection will significantly im prove CAAS when modelling a user’s flow curve (see chap. 5 Model and Architec ture). Here again the ability of next-generation low-price sensors to detect the heart rate and interpret facial expressions will make a difference. Also this is clearly the path game hardware developers are taking as the Kinect of Microsoft’s Xbox One also interprets users’ facial expressions. While the technical implementation of emotion detection has already been solved in the context of affective computing (Geller, 2014), the design of adequate interventions and the development of adequate gamification elements as well as their seamless integration into the workspace will be a huge area for research and development. With respect to the study results several questions for future research have been opened: A long term study is needed to determine if the effects on motivation, ac ceptance, product quality and speed will change over time in an everyday usage situ ation. Another issue that came up in the study (see chap. 7.5 Experiment Results and Discussion) is the high variance between the impaired users. This variance can occur within short periods of times. Thus a future research focus is the development of a game-based assessment tool for impaired users that allows to measure their motoric and cognitive skills quickly and efficiently. Such an assessment tool will help future researchers to “calibrate” the performance index and ensure that changes in an indi vidual user’s performance can safely be attributed to the augmentations of CAAS. This research has already been commenced in the follow-up project. Conclusion: Future Research 165 Finally we wonder to what extent the augmented assistive systems portrayed here and developed in the near future can be used with un-impaired persons. It would be inter esting to adapt CAAS accordingly and quantify the effects on motivation, speed and quality of the work of un-impaired workers. 8.3.2 Ethical Perspectives The ethical implications of CAAS need to be discussed and evaluated. An essential issue is the security of personal data. Especially when sensors are used to detect a worker’s heart rate or emotions it will be necessary to provide more than technical solutions. While the ethical guideline (see chap. 8.2) may be a first step, new ethical rules and standards need to be established and evaluated in pilot projects. This cautious attitude regarding personal data also has to be applied to gamification; while its implementation benefits strongly from the advances mentioned before, the question remains if it is legitimate to replace or supplement intrinsic motivation with extrinsic methods. Here studies of larger groups using CAAS with and without gam ification under the same conditions (skills, workplace, product, payment etc.) are needed. Only if gamification elements in work environments increase the motivation and life satisfaction of a worker in the long run (or if they have no significant influence but increase the productivity) it is legitimate to use them as common feature of CAAS. Generally a model for the early integration of ethical experts and users should be es tablished. Based on the growing societal importance of assistive systems due to de mographic change (see chap. 2.2 Demography and Targeted Users) also the integration of stakeholders from the economic sphere (unions, work councils) should be formalized: as the World Health Organization has stated, design is no longer lim ited to the technological sphere and the user but involves the environment. Within the strongly regulated and confined industrial domain the shift towards “design for all” can be ideally tested and optimized. 166 Supplement: Implementation Details 9 SUPPLEMENT This chapter contains additional information that was not relevant for the main line of argumentation in the text but might be of interest for subsequent studies and imple mentation efforts based on this work. 9.1 Implementation Details To implement the triggers the PrimeSense OpenNI data (see chap. 3.1 Motion Recog nition) had to be accessed using the class VideoFrameGrabber partly developed by the author ‘init1045a’: Equation 8: Video frame grabber function for RGB data. Supplement: Implementation Details 167 The following code snippet (Equation 9) shows how OpenNI and NITE are accessed and how the infrared (IR) and depth information is integrated. Equation 9: OpenNI and NITE configurator function for depth data. The following code snippets (Equation 10, Equation 11) show how a trigger box is set up and how it is checked. Equation 10: Setup of the trigger box. 168 Supplement: Questionnaires in Detail Equation 11: Check of the trigger box. 9.2 Questionnaires in Detail As described in the studies and evaluation chapter (see chap. 7.2.1 Questionnaire De sign) the items in the questionnaires are deliberately redundant and include contradic tory items to detect affirmative tendencies. While the items were thematically sorted in the main text, in the following we provide the original questionnaires where the items are scrambled to make the redundancies and contradictions less obvious. We also provide the corresponding answering data. Table 36: Pre-experiment questions – identical in all scenarios. 1 very good good normal not good bad not at all hardly normal dis tinctly very much not at all <2 hours 5 - 10 > 10 hours hours How are you today? 2 Are you nervous? 3 Do you look forward to the experiment? 4 How many hours per week do you work at the computer? 5 How many hours per week do you play com puter or console games? 2-5 hours 169 Supplement: Questionnaires in Detail Table 37: Post-experiment questions: SotA. fully agree 6 The help device was easy. 7 I would use the help device again. 8 I would recommend the help device to my colleagues. 9 I need somebody’s help to use the help device. 10 The help device was unnecessary complicated. 11 The indicator for picks was easy. 12 I found it tiring to confirm each step with a button. 13 Working with the help device can be learned quickly. 14 I found the work with the help device ex hausting. 15 I could use the help device proficiently. 16 I had to learn a lot before I could work with the help device. 17 I found the instruction on the monitor com plicated. 18 I found the instruction on the monitor easy. 19 I found it tiring to always check the instruc tion on the monitor above me. agree neutral do not agree do not agree at all 170 Supplement: Questionnaires in Detail Question items 1-10 Subject Age Sex Wage level 1 28 0 101 45 1 2 96 3 29 1 91 55 1 87 4 38 1 5 85 36 1 84 6 34 1 7 74 8 48 1 107 41 1 111 9 10 30 1 120 43 1 112 11 63 1 12 115 43 1 121 13 63 1 14 92 15 51 1 108 36 1 108 16 51 0 17 96 51 1 116 18 31 1 19 102 20 33 1 88 Q_it01 Q_it02 Q_it03 Q_it04 Q_it05 Q_it06 Q_it07 Q_it08 Q_it09 Q_it10 4 5 4 3 3 2 1 1 5 5 1 1 4 2 1 2 1 1 3 5 2 1 5 5 3 2 2 2 2 4 3 3 5 3 4 1 4 2 5 5 1 3 5 1 1 2 1 1 4 5 2 3 3 1 1 2 2 2 2 4 2 2 4 2 1 1 1 1 5 4 3 2 3 3 1 2 2 3 4 5 3 4 4 1 1 2 2 2 2 4 3 1 3 3 1 2 2 2 4 4 2 1 3 1 3 2 2 2 4 5 1 2 3 1 1 2 2 2 2 4 2 2 3 1 1 3 3 2 4 4 2 2 3 1 1 2 2 2 4 4 2 2 4 1 1 2 2 2 5 5 2 2 4 1 1 2 1 1 3 5 2 1 5 2 2 2 2 2 4 4 2 3 3 1 1 2 1 2 4 1 2 2 3 2 2 2 2 2 4 4 3 2 3 2 2 1 2 2 2 4 Question items 11-19 Table 38: Post-experiment answers: SotA. Subject Age Sex Wage level 1 28 0 101 45 1 2 96 3 29 1 91 55 1 87 4 38 1 5 85 36 1 84 6 34 1 7 74 8 48 1 107 41 1 111 9 10 30 1 120 43 1 112 11 63 1 12 115 43 1 121 13 63 1 14 92 15 51 1 108 36 1 108 16 51 0 17 96 51 1 116 18 31 1 19 102 20 33 1 88 Q_it11 Q_it12 Q_it13 Q_it14 Q_it15 Q_it16 Q_it17 Q_it18 Q_it19 2 4 4 2 2 4 1 2 4 1 2 5 1 1 5 2 4 5 1 4 3 2 2 5 2 4 4 1 2 2 2 2 4 2 5 2 2 5 2 1 2 5 2 4 4 2 4 4 2 2 3 3 4 4 2 5 5 1 2 5 2 2 4 2 4 4 1 2 3 2 4 4 2 2 5 2 2 3 2 4 5 2 4 4 2 2 4 2 4 4 2 4 2 2 1 5 2 4 4 2 4 4 3 2 4 3 4 4 3 4 4 2 2 4 1 4 4 1 5 2 1 1 4 2 2 4 4 4 1 2 2 3 2 2 2 1 5 4 1 2 4 2 2 4 2 5 5 2 2 4 2 4 4 2 4 3 2 1 4 2 4 5 2 5 4 2 2 4 2 4 4 2 3 2 2 2 3 2 4 2 171 Supplement: Questionnaires in Detail Table 39: Post-experiment questions: Projection. fully agree 6 The help device was easy. 7 I would use the help device again. 8 I would recommend the help device to my colleagues. 9 I need somebody’s help to use the help device. 10 The help device was unnecessary complicated. 11 The instruction left of me on the table dis turbed my work. 12 The indicator for picks was easy. 13 I found the instruction left of me on the table complicated. 14 I found it tiring to confirm each step with a button. 15 Working with the help device can be learned quickly. 16 The model in the center of the workspace helped me. 17 I found the instruction left of me on the table easy. 18 I found the work with the help device ex hausting. 19 I could use the help device proficiently. 20 I had to learn a lot before I could work with the help device. 21 I liked having the instruction directly to my left on the table. 22 The model in the center of the workspace bothered me. agree neutral do not agree do not agree at all 172 Supplement: Questionnaires in Detail Question items 1-10 Subject Age Sex Wage level 1 51 1 75 57 0 2 72 3 35 0 72 29 1 82 4 23 0 5 84 27 0 87 6 33 1 7 89 8 52 0 98 27 0 101 9 10 36 1 108 56 0 111 11 28 1 12 116 27 0 108 13 39 0 14 118 15 52 1 121 34 1 125 16 39 0 17 76 46 1 125 18 56 0 19 114 20 49 0 122 Q_it01 Q_it02 Q_it03 Q_it04 Q_it05 Q_it06 Q_it07 Q_it08 Q_it09 Q_it10 3 3 3 1 1 2 3 4 3 3 2 2 4 1 1 2 2 2 4 5 2 4 4 1 1 2 2 4 4 4 1 2 4 2 1 3 2 2 4 4 2 2 2 1 1 3 2 3 3 4 2 3 4 1 1 2 2 3 4 5 3 3 3 2 1 2 3 2 4 4 2 3 3 4 4 2 2 2 3 4 2 1 4 2 1 2 2 2 4 4 2 2 5 3 1 2 1 2 4 4 2 3 3 2 1 2 2 2 4 4 1 3 3 1 1 2 2 3 2 3 2 3 3 2 1 2 2 2 4 4 2 2 3 3 2 2 1 2 4 4 2 4 4 2 1 2 1 2 4 4 2 2 3 2 2 3 3 3 4 4 2 3 3 1 1 3 2 2 3 4 2 3 3 1 1 1 1 1 4 4 2 3 4 1 1 4 2 2 1 4 2 2 3 1 1 4 2 2 4 4 Question items 11-22 Table 40: Post-experiment answers: Projection. Subject Age Sex Wage level 1 51 1 75 57 0 2 72 3 35 0 72 29 1 82 4 23 0 5 84 27 0 87 6 33 1 7 89 8 52 0 98 27 0 101 9 10 36 1 108 56 0 111 11 28 1 12 116 27 0 108 13 39 0 14 118 15 52 1 121 34 1 125 16 39 0 17 76 46 1 125 18 56 0 19 114 20 49 0 122 Q_it11 Q_it12 Q_it13 Q_it14 Q_it15 Q_it16 Q_it17 Q_it18 Q_it19 Q_it20 Q_it21 Q_it22 3 2 3 3 2 2 3 3 3 3 3 3 5 1 4 3 1 2 2 4 2 4 2 5 4 2 4 4 2 2 2 4 2 4 2 4 4 2 4 2 2 1 2 4 2 2 2 4 4 2 3 3 3 2 2 3 3 3 2 4 2 2 3 5 4 2 2 4 2 3 2 4 4 2 2 2 3 2 3 3 3 2 2 4 4 2 3 2 2 2 3 4 2 4 2 4 5 2 5 4 2 2 2 4 2 4 2 5 4 1 4 3 2 2 2 5 2 4 1 4 5 1 4 2 2 2 2 5 2 4 2 4 4 2 4 3 2 2 2 4 2 5 2 4 4 2 4 4 2 2 2 5 2 2 2 4 5 2 4 4 2 2 2 4 2 4 2 5 4 1 3 2 2 2 2 3 2 2 2 4 4 2 5 4 2 2 2 4 2 4 2 4 3 2 4 4 2 2 2 4 2 4 2 4 5 1 4 4 2 2 2 4 2 3 2 4 4 2 4 2 2 2 2 4 2 4 2 4 4 2 4 4 2 2 2 4 2 2 2 4 173 Supplement: Questionnaires in Detail Table 41: Post-experiment questions: Gamification. fully agree 6 The help device was easy. 7 I would use the help device again. 8 I would recommend the help device to my colleagues. 9 I need somebody’s help to use the help device. 10 The help device was unnecessary complicated. 11 The indicator for picks was easy. 12 I found the instruction on the monitor com plicated. 13 I liked the game on the monitor. 14 I found it tiring to confirm each step with a button. 15 Working with the help device can be learned quickly. 16 The comments on my work motivated me. 17 I found the instruction on the monitor easy. 18 The game on the monitor bothered me. 19 I found the work with the help device ex hausting. 20 I could use the help device proficiently. 21 I would like to always have the game at work. 22 I had to learn a lot before I could work with the help device. 23 The comments on my work disturbed me. 24 I found it tiring to always check the instruc tion on the monitor above me. agree neutral do not agree do not agree at all 174 Supplement: Study Result Details Question items 1-10 Subject Age Sex Wage level 1 31 1 102 2 45 0 72 3 40 1 79 4 52 1 82 5 53 1 84 6 54 1 73 7 26 1 86 8 49 1 125 9 43 0 105 10 29 1 108 11 49 0 116 12 46 1 112 13 47 1 106 14 55 0 110 15 59 1 122 16 37 1 118 17 57 1 124 18 24 1 115 19 45 1 96 20 29 1 91 Q_it01 Q_it02 Q_it03 Q_it04 Q_it05 Q_it06 Q_it07 Q_it08 Q_it09 Q_it10 2 2 3 3 2 2 3 2 5 4 2 2 4 2 2 1 1 2 4 5 2 3 4 2 2 2 3 2 4 4 3 2 3 1 1 2 2 2 2 5 2 3 4 1 1 2 2 2 4 4 2 3 3 1 1 2 2 2 4 4 2 3 4 1 2 1 2 2 4 4 2 2 3 1 1 2 3 2 2 4 2 3 3 1 1 2 4 3 4 4 2 2 3 1 3 2 2 2 4 4 2 4 3 2 2 2 1 1 4 5 2 3 3 4 4 2 3 2 4 3 2 2 4 1 1 2 2 1 4 4 2 3 4 1 1 2 2 1 5 4 2 2 4 1 1 2 2 2 4 4 2 3 3 3 3 2 2 2 3 4 2 3 3 1 1 2 2 3 2 4 2 2 4 1 1 2 3 2 2 4 2 2 3 1 1 2 2 2 4 5 2 2 4 4 4 2 2 2 4 4 Question items 11-24 Table 42: Post-experiment answers: Gamification. Subject Age Sex Wage level 1 31 1 102 2 45 0 72 3 40 1 79 4 52 1 82 5 53 1 84 6 54 1 73 7 26 1 86 8 49 1 125 9 43 0 105 10 29 1 108 11 49 0 116 12 46 1 112 13 47 1 106 14 55 0 110 15 59 1 122 16 37 1 118 17 57 1 124 18 24 1 115 19 45 1 96 20 29 1 91 Q_it11 Q_it12 Q_it13 Q_it14 Q_it15 Q_it16 Q_it17 Q_it18 Q_it19 Q_it20 Q_it21 Q_it22 Q_it23 Q_it24 2 4 2 4 2 2 2 5 4 2 4 4 3 3 2 4 2 4 2 2 2 4 5 2 2 2 5 2 2 4 2 3 2 2 2 5 4 2 2 2 5 3 2 4 2 3 2 1 2 5 4 2 2 3 5 4 2 4 2 3 2 2 2 4 4 2 2 4 4 4 2 4 2 4 3 2 2 4 4 2 2 3 4 4 2 4 1 5 2 1 2 5 4 2 1 4 4 4 2 4 2 4 3 3 2 4 5 2 3 4 4 4 2 2 2 4 3 2 4 4 4 4 3 3 4 4 2 4 2 2 3 2 2 4 3 2 2 2 4 2 2 4 1 4 2 2 2 4 4 1 1 4 5 4 2 4 2 2 2 2 2 4 4 2 2 4 4 4 2 4 2 4 3 1 2 4 4 2 4 4 4 4 2 4 1 4 2 2 2 4 4 2 2 4 4 4 2 4 2 2 2 2 2 5 2 2 2 4 4 2 2 2 2 2 4 2 2 4 3 2 2 2 4 2 2 4 2 2 2 2 2 4 2 3 2 3 2 4 2 4 2 4 2 2 2 4 4 2 4 4 4 4 2 4 2 4 2 2 2 4 4 2 2 4 4 4 2 4 3 4 2 2 2 3 4 2 4 4 4 4 9.3 Study Result Details As indicated in chapter 7.4 Data, even at the most detailed view the data provided is already aggregated in some form – for example sequence completion times are given instead of the process completion times (there are eight processes in each sequence, i.e. 64 processes to analyze for each user). The complete data is provided in the fol lowing. The abbreviation “A” is used for assembly process, the abbreviation “S” for assembly sequence i.e. for eight assembly processes. Since there are very few strong outliers, the graphs are capped at 150 seconds per process. However, the complete data includ ing the outliers can be assessed in the corresponding tables. Process times printed in red indicate that an error occurred in this process. Figure 80: Detailed process times of the SotA population: graph. process time in seconds 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 subject 2 subject 12 subject 1 subject 11 subject 3 subject 13 subject 4 subject 14 subject 5 subject 15 subject 6 subject 16 subject 17 subject 7 subject 18 subject 8 subject 19 subject 9 subject 20 subject 10 A1.1 A1.2 A1.3 A1.4 A1.5 A1.6 A1.7 A1.8 A2.1 A2.2 A2.3 A2.4 A2.5 A2.6 A2.7 A2.8 A3.1 A3.2 A3.3 A3.4 A3.5 A3.6 A3.7 A3.8 A4.1 A4.2 A4.3 A4.4 A4.5 A4.6 A4.7 A4.8 A5.1 A5.2 A5.3 A5.4 A5.5 A5.6 A5.7 A5.8 A6.1 A6.2 A6.3 A6.4 A6.5 A6.6 A6.7 A6.8 A7.1 A7.2 A7.3 A7.4 A7.5 A7.6 A7.7 A7.8 A8.1 A8.2 A8.3 A8.4 A8.5 A8.6 A8.7 A8.8 Supplement: Study Result Details 175 176 Supplement: Study Result Details Sequences 1 and 2, Processes 1-16 Subject Age Sex Wage level 1 28 0 101 45 1 96 2 29 1 91 3 55 1 87 4 38 1 85 5 36 1 84 6 34 1 74 7 48 1 107 8 41 1 111 9 30 1 120 10 43 1 112 11 63 1 115 12 43 1 121 13 63 1 92 14 51 1 108 15 36 1 108 16 51 0 96 17 51 1 116 18 31 1 102 19 20 33 1 88 A1.1 A1.2 A1.3 A1.4 A1.5 A1.6 A1.7 A1.8 61 21 147 81 16 21 31 14 47 39 55 74 58 55 38 42 56 33 38 75 48 51 77 48 45 22 31 21 17 11 14 16 54 36 33 122 63 45 55 44 51 34 33 40 20 15 25 31 77 56 31 56 42 30 31 26 89 27 30 37 54 34 48 32 86 54 60 76 129 65 103 36 39 19 28 21 19 23 17 14 41 31 34 31 19 20 44 30 29 31 31 55 34 62 19 102 56 23 25 83 33 37 24 27 69 62 88 198 51 40 58 49 23 39 67 20 13 44 26 18 46 33 35 61 46 29 72 34 71 52 22 77 45 18 42 33 35 22 33 60 25 39 28 48 50 43 46 41 56 25 26 40 50 29 37 28 21 21 27 40 S1 Errors A2.1 A2.2 A2.3 A2.4 A2.5 A2.6 A2.7 A2.8 392 13% 67 31 42 17 17 8 18 28 408 0% 42 65 35 19 32 23 28 36 425 38% 41 30 44 49 65 34 27 57 177 0% 12 27 12 16 6 8 20 15 452 13% 33 23 13 25 24 17 16 47 249 63% 41 31 22 24 19 27 23 17 347 13% 45 26 36 24 25 35 26 23 351 38% 34 20 35 20 28 19 14 34 608 0% 60 39 37 89 140 37 39 47 180 13% 40 11 13 13 12 11 18 13 250 38% 36 16 20 21 50 25 23 21 361 38% 28 19 14 27 44 23 97 70 307 0% 45 27 22 40 42 52 21 14 614 0% 84 44 38 42 22 30 13 23 250 25% 11 7 9 5 5 10 7 20 356 63% 42 12 15 20 17 19 19 25 23 17 24 10 12 23 15 23 359 50% 290 38% 24 16 24 21 24 22 24 34 20 27 39 39 21 22 24 327 38% 27 50 13 27 21 12 19 26 53 252 0% S2 Errors 228 13% 279 25% 347 38% 116 0% 198 13% 205 63% 239 13% 204 38% 489 0% 131 0% 212 0% 321 25% 264 25% 296 38% 74 25% 169 38% 147 50% 188 13% 217 50% 221 25% Sequences 3 and 4, Processes 17-32 Subject Age Sex Wage level 1 28 0 101 45 1 96 2 29 1 91 3 55 1 87 4 38 1 85 5 36 1 84 6 34 1 74 7 48 1 107 8 41 1 111 9 30 1 120 10 43 1 112 11 63 1 115 12 43 1 121 13 63 1 92 14 51 1 108 15 36 1 108 16 51 0 96 17 51 1 116 18 31 1 102 19 20 33 1 88 A3.1 A3.2 A3.3 A3.4 A3.5 A3.6 A3.7 A3.8 83 21 80 12 11 9 12 30 49 25 33 23 32 20 35 49 39 17 43 25 22 18 25 59 11 8 20 7 6 11 9 11 34 18 22 19 13 22 22 27 21 22 22 18 17 15 53 24 54 16 20 13 13 30 23 25 20 11 33 14 51 15 18 18 56 42 38 39 47 34 20 55 41 14 13 14 16 10 13 10 26 14 28 15 26 11 16 20 25 14 34 28 16 33 29 108 48 19 16 24 19 27 20 35 59 25 47 11 18 24 16 33 19 10 5 6 11 6 6 12 27 16 22 13 11 21 11 21 12 8 15 14 16 18 12 12 97 23 24 65 39 15 15 21 24 18 22 22 21 17 26 42 55 15 23 13 10 15 15 39 S3 Errors A4.1 A4.2 A4.3 A4.4 A4.5 A4.6 A4.7 A4.8 258 0% 76 15 29 9 12 13 11 28 265 25% 27 49 44 27 31 22 25 23 248 13% 38 16 19 32 62 26 24 46 84 25% 6 5 25 8 6 5 7 11 176 63% 43 25 18 22 15 12 47 17 192 63% 23 16 22 14 27 19 29 22 194 13% 66 15 31 26 16 14 21 28 180 38% 17 8 17 8 30 11 36 9 331 0% 54 61 43 41 57 28 55 48 131 13% 17 7 8 10 30 22 10 8 157 0% 26 19 22 11 17 12 12 11 287 38% 47 14 20 21 26 22 24 40 208 13% 19 18 17 12 11 15 12 28 233 0% 67 27 16 11 16 43 16 21 76 0% 8 5 3 4 5 3 10 15 140 50% 30 9 12 10 16 14 28 43 108 63% 18 13 9 12 9 26 12 20 299 0% 56 28 15 16 18 17 16 42 193 13% 38 22 29 20 19 12 25 24 184 0% 29 23 16 8 9 11 12 56 S4 Errors 192 0% 247 50% 263 13% 73 0% 199 13% 172 63% 217 13% 137 38% 387 0% 112 38% 128 0% 214 13% 132 25% 216 13% 53 25% 161 75% 120 38% 207 0% 187 25% 164 25% Sequences 5 and 6, Processes 33-48 Subject Age Sex Wage level 1 28 0 101 45 1 96 2 29 1 91 3 55 1 87 4 38 1 85 5 36 1 84 6 34 1 74 7 48 1 107 8 41 1 111 9 30 1 120 10 43 1 112 11 63 1 115 12 43 1 121 13 63 1 92 14 51 1 108 15 36 1 108 16 51 0 96 17 51 1 116 18 31 1 102 19 20 33 1 88 A5.1 A5.2 A5.3 A5.4 A5.5 A5.6 A5.7 A5.8 26 19 22 10 8 16 15 13 19 42 35 34 32 30 37 37 40 18 37 29 31 22 23 32 9 13 8 7 6 7 6 13 25 26 16 12 13 15 25 26 24 14 16 10 13 22 18 27 37 15 17 13 18 9 27 31 20 12 25 13 20 16 12 38 59 30 35 58 66 54 45 29 12 17 12 9 11 8 7 9 11 30 36 14 61 39 13 19 24 14 12 13 80 29 23 23 34 17 30 12 15 11 14 16 35 35 15 28 10 17 17 41 8 3 6 6 4 4 17 18 14 8 10 13 6 8 66 19 14 9 12 12 10 12 19 20 20 26 19 17 20 9 17 140 29 14 13 18 12 13 17 21 8 12 11 18 28 12 15 17 S5 Errors A6.1 A6.2 A6.3 A6.4 A6.5 A6.6 A6.7 A6.8 129 0% 22 28 17 9 8 13 14 20 266 50% 37 33 23 26 30 21 24 28 231 13% 32 19 24 34 22 28 17 36 70 25% 7 9 10 8 5 7 7 19 158 13% 22 18 14 12 13 22 21 27 143 63% 17 12 10 11 8 9 41 32 167 13% 26 12 34 14 10 11 51 42 155 0% 13 12 13 9 33 35 22 47 376 0% 46 26 23 39 26 43 39 30 84 13% 20 9 11 11 13 11 9 9 222 0% 20 11 16 12 16 9 12 32 218 38% 19 11 14 14 28 18 26 26 149 13% 38 13 10 10 10 16 23 11 198 13% 25 32 24 36 37 14 27 32 66 25% 11 2 5 3 3 3 10 17 144 75% 15 8 14 12 7 7 10 11 106 63% 8 15 9 11 12 21 16 16 26 268 13% 44 16 18 21 15 10 27 136 25% 18 39 27 19 18 16 14 20 15 10 11 16 120 13% 14 10 13 8 S6 Errors 131 25% 221 25% 213 13% 73 0% 149 13% 139 63% 200 13% 183 38% 272 0% 94 13% 127 13% 156 25% 129 13% 227 0% 54 25% 84 50% 107 63% 177 13% 171 50% 97 13% Sequences 7 and 8, Processes 49-64 Table 43: Detailed process times of the SotA population: values. Subject Age Sex Wage level 1 28 0 101 45 1 96 2 29 1 91 3 55 1 87 4 38 1 85 5 36 1 84 6 34 1 74 7 48 1 107 8 41 1 111 9 30 1 120 10 43 1 112 11 63 1 115 12 43 1 121 13 63 1 92 14 51 1 108 15 36 1 108 16 51 0 96 17 51 1 116 18 31 1 102 19 20 33 1 88 A7.1 A7.2 A7.3 A7.4 A7.5 A7.6 A7.7 A7.8 25 12 12 11 5 8 14 16 39 33 34 25 15 23 26 31 28 18 14 28 17 16 17 46 11 12 8 8 5 6 7 10 23 20 10 17 18 14 11 17 24 11 10 11 11 9 13 24 40 9 28 11 9 15 15 20 22 14 16 9 13 13 14 33 37 23 31 43 39 26 43 28 15 10 13 11 9 8 7 9 23 10 14 10 35 13 13 14 20 14 11 11 19 18 20 26 27 8 15 8 9 14 21 15 20 23 20 21 23 12 16 34 9 3 5 3 2 3 13 18 16 11 7 7 11 16 8 13 15 16 16 12 10 10 12 10 13 13 15 20 28 25 27 14 15 9 22 17 16 17 16 21 14 9 7 10 6 8 7 18 S7 Errors A8.1 A8.2 A8.3 A8.4 A8.5 A8.6 A8.7 A8.8 17 6 8 10 13 103 25% 17 18 11 15 24 22 30 32 226 0% 38 23 34 47 10 22 18 17 184 13% 30 20 19 10 5 5 10 11 67 0% 12 11 15 10 14 14 12 19 131 13% 17 14 15 10 10 14 11 18 113 63% 16 12 8 11 9 15 26 30 147 13% 36 12 27 13 13 11 13 21 134 38% 21 23 13 57 44 25 56 20 269 0% 36 30 38 6 5 7 7 13 84 13% 16 12 8 13 21 9 15 40 132 25% 27 12 16 12 28 13 37 44 139 25% 20 11 11 14 8 9 16 10 116 13% 25 14 12 16 10 16 12 30 169 25% 19 18 9 3 4 7 6 14 56 25% 9 4 4 7 8 20 16 10 90 63% 28 5 5 101 50% 23 9 11 12 10 12 22 12 153 0% 23 12 8 12 12 15 16 24 22 19 24 13 10 17 13 13 133 25% 79 13% 16 11 17 21 11 8 8 16 S8 Errors 100 25% 217 25% 183 13% 79 0% 114 13% 98 63% 166 13% 128 0% 306 0% 75 13% 153 0% 176 38% 108 25% 130 0% 51 0% 99 63% 110 63% 123 13% 130 25% 107 13% Figure 81: Detailed process times of the Projection population: graph. process time in seconds 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 subject 2 subject 12 subject 1 subject 11 subject 13 subject 3 subject 14 subject 4 subject 15 subject 5 subject 16 subject 6 subject 17 subject 7 subject 8 subject 18 subject 9 subject 19 subject 10 subject 20 A1.1 A1.2 A1.3 A1.4 A1.5 A1.6 A1.7 A1.8 A2.1 A2.2 A2.3 A2.4 A2.5 A2.6 A2.7 A2.8 A3.1 A3.2 A3.3 A3.4 A3.5 A3.6 A3.7 A3.8 A4.1 A4.2 A4.3 A4.4 A4.5 A4.6 A4.7 A4.8 A5.1 A5.2 A5.3 A5.4 A5.5 A5.6 A5.7 A5.8 A6.1 A6.2 A6.3 A6.4 A6.5 A6.6 A6.7 A6.8 A7.1 A7.2 A7.3 A7.4 A7.5 A7.6 A7.7 A7.8 A8.1 A8.2 A8.3 A8.4 A8.5 A8.6 A8.7 A8.8 Supplement: Study Result Details 177 178 Supplement: Study Result Details Sequences 1 and 2, Processes 1-16 Subject Age Sex Wage level 1 51 1 75 57 0 72 2 3 35 0 72 29 1 82 4 23 0 84 5 27 0 87 6 33 1 89 7 52 0 98 8 27 0 101 9 36 1 108 10 56 0 111 11 28 1 116 12 27 0 108 13 39 0 118 14 52 1 121 15 34 1 125 16 39 0 76 17 46 1 125 18 56 0 114 19 20 49 0 122 A1.1 A1.2 A1.3 A1.4 A1.5 A1.6 A1.7 A1.8 S1 88 36 89 115 47 36 48 41 500 85 47 59 61 41 28 25 30 377 52 41 25 41 191 27 26 49 450 33 59 60 41 21 12 49 39 314 73 74 52 27 34 37 61 30 389 87 33 45 34 39 42 31 44 354 28 24 28 31 26 11 23 30 200 57 20 24 86 20 22 23 16 267 44 41 31 27 25 20 89 45 323 65 13 28 31 19 15 21 13 204 73 45 49 47 46 29 66 21 377 17 19 14 15 12 10 14 9 110 56 32 29 75 28 26 42 20 307 33 40 36 47 38 41 36 24 293 42 28 64 22 309 52 34 35 34 17 15 15 13 11 18 13 14 117 35 63 161 76 39 37 20 21 452 27 17 23 56 30 23 30 64 271 49 114 22 21 60 18 40 32 355 57 22 73 44 21 16 38 14 283 Sequences 3 and 4, Processes 17-32 Subject Age Sex Wage level 1 51 1 75 57 0 72 2 3 35 0 72 29 1 82 4 23 0 84 5 27 0 87 6 33 1 89 7 52 0 98 8 27 0 101 9 36 1 108 10 56 0 111 11 28 1 116 12 27 0 108 13 39 0 118 14 52 1 121 15 34 1 125 16 39 0 76 17 46 1 125 18 56 0 114 19 20 49 0 122 A3.1 A3.2 A3.3 A3.4 A3.5 A3.6 A3.7 A3.8 S3 Errors A4.1 A4.2 A4.3 A4.4 A4.5 A4.6 A4.7 A4.8 S4 Errors 105 31 57 80 44 37 27 33 413 13% 60 27 70 74 88 46 220 30 616 25% 48 36 41 30 36 40 51 29 312 63% 33 17 19 24 19 21 65 22 220 63% 19 9 47 38 15 20 23 18 189 88% 11 14 12 43 16 21 11 8 135 100% 36 16 20 21 14 14 15 14 149 0% 56 17 21 13 17 13 15 20 172 0% 41 73 33 22 24 42 40 18 292 13% 50 32 31 22 14 24 30 23 225 13% 37 15 21 34 16 25 14 26 187 63% 33 16 16 20 17 14 18 42 177 63% 21 31 13 12 12 16 15 21 140 13% 32 14 19 8 9 9 10 19 120 13% 33 12 11 8 10 10 9 9 103 0% 19 9 10 9 9 7 9 14 86 0% 24 13 15 16 11 7 14 22 122 13% 9 16 16 43 14 6 15 18 137 13% 23 18 18 11 9 9 13 18 119 0% 53 9 17 10 10 11 16 14 140 0% 21 34 13 22 16 15 28 15 164 13% 23 20 15 10 12 10 11 11 111 13% 11 5 9 6 7 5 8 4 55 0% 8 4 7 5 5 6 7 5 47 0% 37 15 14 60 13 10 10 18 177 13% 30 11 10 15 13 11 13 20 122 0% 18 15 11 28 22 13 12 21 141 25% 18 11 12 12 16 20 13 12 114 13% 51 16 46 23 33 27 37 37 271 13% 68 25 37 21 31 22 17 15 236 13% 12 8 7 11 8 17 6 10 77 13% 9 7 9 11 7 25 8 11 88 13% 19 19 23 27 43 41 24 31 227 0% 19 15 16 40 48 55 55 50 299 63% 33 10 15 17 20 17 13 12 137 13% 34 9 15 12 19 15 12 12 128 13% 28 11 13 36 16 14 10 38 167 100% 27 15 33 11 37 40 74 22 258 25% 19 9 10 17 8 10 16 20 108 38% 11 16 8 10 9 5 12 8 80 38% Sequences 5 and 6, Processes 33-48 Subject Age Sex Wage level 1 51 1 75 57 0 72 2 3 35 0 72 29 1 82 4 23 0 84 5 27 0 87 6 33 1 89 7 52 0 98 8 27 0 101 9 36 1 108 10 56 0 111 11 28 1 116 12 27 0 108 13 39 0 118 14 52 1 121 15 34 1 125 16 39 0 76 17 46 1 125 18 56 0 114 19 20 49 0 122 A5.1 A5.2 A5.3 A5.4 A5.5 A5.6 A5.7 A5.8 S5 Errors A6.1 A6.2 A6.3 A6.4 A6.5 A6.6 A6.7 A6.8 S6 Errors 42 21 87 66 43 93 43 31 426 50% 47 48 55 47 75 45 30 29 377 25% 24 15 14 22 14 22 35 29 174 75% 35 14 19 23 11 17 17 31 168 63% 13 12 7 14 24 82 20 9 181 100% 12 11 12 11 16 14 10 19 105 100% 33 11 18 21 11 11 19 14 138 0% 18 21 10 10 12 9 12 13 106 0% 30 19 23 12 13 16 21 16 150 13% 41 23 40 19 16 20 20 23 202 25% 28 15 15 24 16 14 20 27 158 63% 35 16 16 42 18 17 18 20 182 63% 29 10 9 11 12 10 9 13 102 13% 16 12 11 8 9 11 8 13 88 13% 12 10 8 12 6 8 7 9 72 0% 18 10 10 11 6 6 9 7 77 0% 11 19 10 14 9 9 14 22 107 13% 14 13 8 10 10 7 9 23 94 13% 26 10 14 10 12 9 24 17 122 0% 22 13 12 12 9 8 13 15 105 0% 19 15 9 8 12 10 28 12 114 13% 19 14 12 11 11 7 12 10 96 13% 9 4 5 4 4 4 5 5 40 0% 9 5 6 5 4 5 6 8 46 0% 11 9 9 11 9 11 9 15 85 0% 24 9 13 9 11 9 16 15 106 0% 19 9 13 18 15 11 14 21 120 13% 21 15 24 13 9 11 11 20 122 13% 59 26 30 23 21 16 20 19 213 13% 53 23 28 17 20 16 29 19 204 13% 6 6 9 9 7 9 9 7 62 13% 8 8 8 10 8 11 7 11 70 13% 27 14 12 32 14 10 10 52 172 75% 26 12 12 47 36 14 35 21 204 88% 28 12 14 11 16 13 10 16 120 13% 21 12 13 10 12 12 12 11 102 13% 20 9 8 8 9 18 114 31 217 50% 24 6 15 13 15 13 24 13 122 100% 20 8 8 12 13 6 15 37 118 100% 29 8 9 14 9 26 14 10 120 88% Sequences 7 and 8, Processes 49-64 Table 44: Detailed process times of the Projection population: values. Subject Age Sex Wage level 1 51 1 75 57 0 72 2 3 35 0 72 29 1 82 4 23 0 84 5 27 0 87 6 33 1 89 7 52 0 98 8 27 0 101 9 36 1 108 10 56 0 111 11 28 1 116 12 27 0 108 13 39 0 118 14 52 1 121 15 34 1 125 16 39 0 76 17 46 1 125 18 56 0 114 19 20 49 0 122 A7.1 A7.2 A7.3 A7.4 A7.5 A7.6 A7.7 A7.8 S7 Errors A8.1 A8.2 A8.3 A8.4 A8.5 A8.6 A8.7 A8.8 S8 Errors 46 20 71 36 31 41 66 47 357 50% 57 33 28 32 47 39 26 24 286 25% 22 15 10 13 12 17 25 39 153 63% 17 13 11 9 9 15 9 16 99 63% 23 7 6 23 9 57 13 8 147 100% 16 8 7 25 13 25 10 35 139 100% 21 10 35 14 14 11 12 11 127 0% 23 9 9 10 11 11 15 8 95 0% 24 21 20 11 15 16 20 19 146 25% 48 16 13 15 13 13 20 17 156 25% 44 14 12 29 15 14 22 29 179 63% 36 18 17 17 13 15 17 18 149 63% 25 12 12 14 9 8 12 11 104 13% 19 12 11 10 10 9 14 10 94 13% 20 13 11 9 6 9 10 6 83 13% 14 8 13 7 8 5 7 13 75 13% 15 15 16 9 7 8 11 13 93 13% 18 6 15 10 6 10 6 14 85 13% 45 10 10 18 11 8 12 11 125 0% 18 8 9 9 16 8 11 12 91 0% 17 16 9 11 14 9 12 10 98 13% 16 15 13 12 9 8 17 12 102 13% 8 5 5 6 4 5 6 5 42 0% 9 4 5 9 3 4 6 4 43 0% 25 7 14 10 9 9 13 28 115 13% 12 7 8 9 5 7 9 23 80 0% 11 9 11 10 11 8 11 16 86 13% 12 9 6 10 8 9 13 11 77 13% 50 22 29 16 23 15 17 17 187 13% 50 19 28 20 19 21 15 16 187 13% 6 7 11 9 15 7 7 10 73 13% 7 7 10 10 5 8 17 10 74 0% 20 21 19 41 25 29 229 75% 16 20 21 24 16 20 12 30 158 75% 63 13 27 10 12 9 9 11 9 11 97 13% 21 12 10 8 8 10 9 15 93 13% 30 9 17 17 17 17 14 42 164 63% 16 16 5 27 7 12 27 17 127 63% 22 10 10 20 11 9 11 8 102 38% 11 18 11 12 5 5 10 11 84 38% Errors A2.1 A2.2 A2.3 A2.4 A2.5 A2.6 A2.7 A2.8 S2 Errors 13% 59 23 40 54 51 53 24 28 331 13% 13% 68 47 47 53 39 22 20 33 329 38% 75% 21 13 37 22 16 47 56 15 226 88% 100% 66 22 26 22 28 26 21 17 227 13% 50% 100 32 130 32 14 35 18 20 380 13% 63% 44 27 65 47 25 20 21 51 299 63% 13% 18 14 12 31 14 17 21 20 145 0% 0% 32 13 9 13 13 13 15 10 118 0% 100% 37 16 21 31 15 18 17 20 175 38% 0% 47 12 16 13 11 10 15 17 141 0% 13% 60 19 21 21 24 13 20 14 192 13% 0% 17 6 10 7 7 6 6 7 66 0% 25% 44 19 15 67 13 18 15 25 215 25% 13% 33 20 22 19 46 15 23 35 213 13% 25% 96 21 68 41 39 19 19 30 332 13% 0% 27 10 9 15 14 8 11 14 108 13% 100% 33 17 39 74 58 34 28 29 313 63% 63% 30 13 51 46 55 21 14 11 240 13% 25% 33 20 12 39 21 15 29 36 205 100% 88% 32 12 27 23 18 12 17 22 162 38% Figure 82: Detailed process times of the Gamification population: graph. process time in seconds 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 subject 2 subject 12 subject 1 subject 11 subject 3 subject 13 subject 4 subject 14 subject 5 subject 15 subject 6 subject 16 subject 17 subject 7 subject 18 subject 8 subject 19 subject 9 subject 20 subject 10 A1.1 A1.2 A1.3 A1.4 A1.5 A1.6 A1.7 A1.8 A2.1 A2.2 A2.3 A2.4 A2.5 A2.6 A2.7 A2.8 A3.1 A3.2 A3.3 A3.4 A3.5 A3.6 A3.7 A3.8 A4.1 A4.2 A4.3 A4.4 A4.5 A4.6 A4.7 A4.8 A5.1 A5.2 A5.3 A5.4 A5.5 A5.6 A5.7 A5.8 A6.1 A6.2 A6.3 A6.4 A6.5 A6.6 A6.7 A6.8 A7.1 A7.2 A7.3 A7.4 A7.5 A7.6 A7.7 A7.8 A8.1 A8.2 A8.3 A8.4 A8.5 A8.6 A8.7 A8.8 Supplement: Study Result Details 179 180 Supplement: Study Result Details Sequences 1 and 2, Processes 1-16 Subject Age Sex Wage level 1 31 1 102 2 45 0 72 3 40 1 79 4 52 1 82 5 53 1 84 6 54 1 73 7 26 1 86 8 49 1 125 9 43 0 105 10 29 1 108 11 49 0 116 12 46 1 112 13 47 1 106 14 55 0 110 15 59 1 122 16 37 1 118 17 57 1 124 18 24 1 115 19 45 1 96 20 29 1 91 A1.1 A1.2 A1.3 A1.4 A1.5 A1.6 A1.7 A1.8 S1 51 18 28 13 15 33 36 18 211 17 17 37 52 34 55 35 19 266 37 62 32 26 40 31 33 21 281 61 27 33 35 26 22 75 29 308 30 26 22 16 25 48 16 30 212 46 55 59 33 96 102 74 44 509 33 40 28 29 29 19 31 34 243 36 34 27 30 30 22 23 9 211 37 36 41 67 42 27 39 21 310 37 35 17 23 25 15 63 14 228 19 38 25 40 30 25 25 33 236 40 24 38 37 25 22 32 23 241 23 20 20 35 21 26 21 32 199 54 70 83 58 35 39 61 59 457 33 68 25 40 36 23 35 43 303 62 74 43 52 33 28 69 24 385 19 17 18 17 63 41 19 15 208 33 40 55 85 48 49 25 20 355 52 58 37 32 40 34 40 34 328 37 31 29 42 32 36 62 45 314 Sequences 3 and 4, Processes 17-32 Subject Age Sex Wage level 1 31 1 102 2 45 0 72 3 40 1 79 4 52 1 82 5 53 1 84 6 54 1 73 7 26 1 86 8 49 1 125 9 43 0 105 10 29 1 108 11 49 0 116 12 46 1 112 13 47 1 106 14 55 0 110 15 59 1 122 16 37 1 118 17 57 1 124 18 24 1 115 19 45 1 96 20 29 1 91 A3.1 A3.2 A3.3 A3.4 A3.5 A3.6 A3.7 A3.8 S3 Errors A4.1 A4.2 A4.3 A4.4 A4.5 A4.6 A4.7 A4.8 S4 Errors 42 11 13 15 12 7 11 18 128 13% 24 13 17 22 14 11 10 15 125 13% 30 9 13 8 10 16 20 14 120 25% 18 9 7 8 7 12 16 26 103 25% 24 19 18 18 27 21 22 27 175 13% 33 15 18 21 17 20 17 23 164 25% 27 18 29 26 26 17 19 27 188 63% 17 19 20 22 20 16 30 20 164 63% 23 9 11 8 15 40 13 18 137 88% 20 7 8 8 17 19 19 17 113 88% 38 31 27 21 41 37 30 31 256 25% 43 24 27 35 22 25 39 40 254 25% 27 16 14 19 11 11 14 18 131 38% 18 14 13 18 10 16 33 15 135 38% 22 4 11 9 9 20 12 22 108 13% 20 6 8 11 11 8 8 11 84 25% 51 17 20 19 20 14 19 13 172 50% 34 16 25 20 20 15 18 16 162 38% 25 23 18 14 15 27 18 11 151 25% 32 13 11 16 24 12 12 13 131 13% 16 10 11 7 7 8 11 11 82 13% 19 11 8 11 9 7 8 8 81 13% 27 17 22 21 32 27 15 18 179 0% 41 18 24 43 24 14 14 12 190 0% 27 14 15 15 23 15 14 16 136 25% 15 21 11 10 11 26 11 29 133 50% 35 20 22 36 35 30 25 19 223 50% 22 13 24 18 18 10 11 16 132 63% 27 37 16 12 14 23 27 19 175 38% 22 55 11 11 11 17 22 16 166 38% 32 66 31 32 26 28 23 11 248 63% 23 17 16 17 25 15 13 9 134 63% 30 20 17 9 31 15 13 11 145 50% 22 8 15 10 12 8 8 24 107 38% 23 9 28 12 29 17 13 12 144 13% 24 4 33 18 25 27 34 25 191 38% 47 41 36 30 31 37 42 28 292 0% 44 28 48 33 25 30 31 40 279 0% 52 15 23 23 27 18 18 12 187 25% 33 18 21 53 22 18 18 18 200 25% Sequences 5 and 6, Processes 33-48 Subject Age Sex Wage level 1 31 1 102 2 45 0 72 3 40 1 79 4 52 1 82 5 53 1 84 6 54 1 73 7 26 1 86 8 49 1 125 9 43 0 105 10 29 1 108 11 49 0 116 12 46 1 112 13 47 1 106 14 55 0 110 15 59 1 122 16 37 1 118 17 57 1 124 18 24 1 115 19 45 1 96 20 29 1 91 A5.1 A5.2 A5.3 A5.4 A5.5 A5.6 A5.7 A5.8 S5 Errors A6.1 A6.2 A6.3 A6.4 A6.5 A6.6 A6.7 A6.8 S6 Errors 19 11 14 11 16 8 13 16 109 0% 21 9 19 13 11 11 16 15 115 25% 12 12 10 9 8 7 13 10 79 0% 23 6 8 8 11 12 14 9 89 25% 24 13 19 13 16 13 13 31 142 13% 26 17 37 16 14 17 14 30 171 13% 15 18 14 31 16 12 20 9 134 63% 22 13 19 17 18 17 16 26 145 63% 10 11 8 8 7 24 55 30 153 100% 15 6 6 6 14 14 9 58 128 100% 24 24 29 23 34 23 25 26 207 25% 31 25 29 18 23 23 23 30 202 25% 25 15 12 20 10 13 18 14 127 38% 18 11 11 13 10 11 19 21 114 38% 22 4 8 8 6 6 11 11 77 25% 21 17 6 11 15 14 11 13 107 50% 22 27 22 15 16 16 23 22 163 50% 35 19 13 18 19 14 31 13 161 38% 23 21 8 9 11 8 9 10 98 13% 23 29 9 10 10 8 20 11 119 13% 14 14 10 6 8 8 9 9 79 13% 19 7 7 7 6 7 10 9 71 13% 25 13 16 19 17 15 16 13 133 0% 21 12 9 16 15 12 14 14 114 25% 15 11 9 13 12 12 11 21 103 38% 15 37 8 13 11 12 10 19 124 38% 22 10 19 12 8 15 14 15 115 63% 31 14 9 15 15 11 11 15 121 63% 27 10 10 9 8 14 15 21 113 50% 41 11 10 9 9 11 11 14 116 50% 21 4 30 23 22 20 16 20 155 75% 25 7 14 16 27 19 11 7 125 63% 17 21 9 11 15 12 9 12 104 50% 11 7 43 8 15 12 9 11 115 50% 38 15 28 28 24 24 32 17 206 38% 32 10 27 19 17 26 23 14 167 13% 45 36 59 38 40 42 22 31 313 25% 56 22 49 30 38 37 39 40 309 0% 19 15 17 27 16 11 17 14 135 50% 23 12 15 17 11 10 23 22 133 38% Sequences 7 and 8, Processes 49-64 Table 45: Detailed process times of the Gamification population: values. Errors A2.1 A2.2 A2.3 A2.4 A2.5 A2.6 A2.7 A2.8 S2 0% 22 10 15 18 16 11 15 17 124 0% 30 11 11 9 12 15 12 30 131 13% 40 29 17 22 28 22 18 26 202 63% 20 16 14 22 29 34 29 18 181 88% 22 11 15 11 27 13 12 69 179 63% 29 28 42 29 42 30 43 29 272 25% 24 18 15 19 13 15 18 14 136 13% 43 1 11 14 11 18 17 9 124 38% 49 21 21 38 39 34 27 14 242 13% 24 30 26 26 13 17 18 19 173 13% 21 19 16 21 13 13 10 9 123 0% 37 15 25 25 42 17 24 19 205 0% 27 30 31 28 31 18 15 14 193 50% 35 39 43 40 67 31 57 28 340 13% 56 33 58 11 11 16 37 19 241 50% 37 23 25 30 37 33 42 71 299 63% 22 12 23 11 18 19 19 13 137 13% 31 19 29 31 35 30 25 22 221 13% 44 34 27 40 32 43 38 30 288 50% 43 15 29 49 25 21 22 25 229 Errors 38% 25% 25% 63% 75% 25% 25% 13% 38% 13% 13% 0% 13% 13% 25% 63% 50% 50% 0% 25% Subject Age Sex Wage level 1 31 1 102 2 45 0 72 3 40 1 79 4 52 1 82 5 53 1 84 6 54 1 73 7 26 1 86 8 49 1 125 9 43 0 105 10 29 1 108 11 49 0 116 12 46 1 112 13 47 1 106 14 55 0 110 15 59 1 122 16 37 1 118 17 57 1 124 18 24 1 115 19 45 1 96 20 29 1 91 A7.1 A7.2 A7.3 A7.4 A7.5 A7.6 A7.7 A7.8 S7 Errors A8.1 A8.2 A8.3 A8.4 A8.5 A8.6 A8.7 A8.8 S8 Errors 22 6 14 13 8 9 14 13 98 13% 17 9 31 15 11 9 16 12 119 13% 21 7 8 9 8 8 10 13 82 25% 22 9 8 7 6 8 12 8 80 0% 27 15 10 10 16 14 15 16 123 13% 27 20 16 13 10 15 25 15 142 38% 14 6 20 16 11 19 17 14 117 63% 16 13 16 13 16 15 13 12 113 63% 20 6 8 8 12 10 42 7 112 100% 19 7 6 7 9 13 59 55 175 100% 28 21 24 25 26 23 17 15 178 0% 34 21 17 19 22 26 16 18 174 25% 22 13 11 14 13 13 11 15 111 38% 23 12 10 14 13 11 19 12 114 38% 21 13 10 7 8 14 9 11 94 25% 16 4 11 11 16 6 10 4 77 13% 39 16 18 20 14 17 17 15 157 38% 23 9 19 26 15 11 14 13 129 50% 20 9 9 9 9 7 12 9 82 0% 17 16 8 12 10 6 8 9 86 13% 15 10 8 6 11 6 8 9 74 13% 14 8 6 8 8 6 7 7 65 13% 33 16 12 18 13 13 13 9 126 0% 23 13 12 11 12 14 12 13 109 25% 14 20 10 12 8 11 12 15 101 50% 11 12 9 17 15 12 12 17 105 13% 23 8 17 9 13 11 8 18 107 63% 22 8 11 16 10 12 11 12 102 63% 11 6 7 8 8 7 30 16 92 50% 13 11 11 7 9 6 22 10 88 50% 17 12 16 12 16 10 10 16 108 63% 17 13 14 33 20 22 4 8 131 63% 20 13 9 7 12 24 13 10 107 50% 22 8 12 8 13 11 11 10 94 50% 52 12 24 17 26 17 16 16 179 13% 33 10 20 16 20 15 21 11 146 13% 53 32 60 39 58 59 65 53 418 50% 53 35 50 33 34 35 39 38 317 0% 34 15 19 51 18 11 27 15 190 38% 23 9 14 13 12 21 15 21 128 38% References 181 10 REFERENCES English translations of German titles in [square brackets]. AAL Contents Working Group, Task Force. (2013). ICT-based Solutions for Support ing Occupation in Life of Older Adults. Retrieved from http://www.aal-eu rope.eu/wp-content/uploads/2013/03/AAL-2013-6-call-text-20130326.pdf Anders, T. R., Fozard, J. L., & Lillyquist, T. D. (1972). Effects of age upon retrieval from short-term memory. Developmental Psychology, 6(2), 214–217. doi:10.1037/h0032103 Andresen, E. M., Fitch, C. A., McLendon, P. M., & Meyers, A. R. (2000). Reliability and validity of disability questions for US Census 2000. American Journal of Public Health, 90(8), 1297. Retrieved from http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1446332/ Australian Institute of Health and Welfare. (2004). Disability and its relationship to health conditions and other factors. Canberra: Australian Institute of Health and Welfare. Bailey, R. W. (1989). Human performance engineering : using human factors/ergo nomics to achieve computer system usability. Englewood Cliffs, N.J.: Pren tice Hall. Ballard, C. (2011, June 7). Developing for Kinect using open source APIs. Tribal Labs. Retrieved from http://www.triballabs.net/2011/06/kinectapis Bartle, R. (1996). Hearts, Clubs, Diamonds, Spades: Players Who suit MUDs. Re trieved October 10, 2013, from http://www.mud.co.uk/richard/hcds.htm Bierkandt, J., Preissner, M., Hermann, F., & Hipp, C. (2011). Usability und humanmachine interfaces in der Produktion. Studie Qualitätsmerkmale für Ent wicklungswerkzeuge. (D. Spath & A. Weisbecker, Eds.). Stuttgart: Fraun hofer-Verl. Biundo, S., Bidot, J., & Schattenberg, B. (2011). Planning in the Real World. Informatik-Spektrum, 34(5), 443–454. doi:10.1007/s00287-011-0562-7 Brach, M., & Korn, O. (2012). Assistive technologies at home and in the workplace— a field of research for exercise science and human movement science. Eu ropean Review of Aging and Physical Activity, 9(1), 1–4. doi:10.1007/s11556-012-0099-z Brand, S., & Holsboer-Trachsler, E. (2010). Das Burnout Syndrom – eine Übersicht. Therapeutische Umschau, 67(11), 561–565. doi:10.1024/0040 5930/a000095 Brumels, K. A., Blasius, T., Cortright, T., Oumedian, D., & Solberg, B. (2008). Com parison of efficacy between traditional and video game based balance pro grams. Clin Kinesiol, 62(4), 26–31. Retrieved from http://clinicalkinesiology.org/content/journals/2008/win ter/Brumels_et_al_62_4_26-32/index_files/Brumels_et_al_62_4_26 32b.pdf 182 References Buckland, M. (2005). Programming game AI by example. Plano, Texas, USA: Wordware Pub. Burckhardt, J. (1999). The Greeks and Greek civilization. (O. Murray & S. Stern, Eds.). New York: St. Martin’s Press. Card, S. K., Moran, T. P., & Newell, A. (1983). The psychology of human-computer interaction. Hillsdale, N.J.: L. Erlbaum Associates. Cardoso Machado, N. M. (2011). Karl Polanyi and the New Economic Sociology: Notes on the Concept of (Dis)embeddedness. MPRA Paper. Retrieved Sep tember 17, 2013, from http://mpra.ub.uni-muenchen.de/48957/ Castellani, S., Hanrahan, B., & Colombino, T. (2013). Game Mechanics in Support of Production Environments. In CHI ’13 Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems. Retrieved from http://gamification-research.org/wp-content/uploads/2013/03/Castel lani_etal.pdf Charles, D., Kerr, A., McNeill, M., McAlister, M., Black, M., Kcklich, J., … Stringer, K. (2005). Player-centred game design: Player modelling and adaptive dig ital games. In Proceedings of the Digital Games Research Conference (Vol. 285). Retrieved from http://www.researchgate.net/publica tion/228636408_Player-centred_game_design_Player_model ling_and_adaptive_digital_games/file/9fcfd514853cb27e1f.pdf Cook, A. M. (2010). The future of assistive technologies. In ASSETS ’10 Proceedings of the 12th international ACM SIGACCESS conference on Computers and accessibility (p. 1). ACM Press. doi:10.1145/1878803.1878805 Cook, A. M., & Hussey, S. M. (1995). Assistive technologies : principles and practice. St. Louis: Mosby. Cowie, P. (2013, September 23). The Phenomena of Gamification – The Next Big Thing for Employers? Retrieved November 5, 2013, from http://www.en terprise-gamification.com/index.php?option=com_content&view=arti cle&id=167:the-phenomena-of-gamification-the-next-big-thing-for employers&catid=4:blog&Itemid=251&lang=de Csíkszentmihályi, M. (1975). Beyond boredom and anxiety. San Francisco, USA: Jossey-Bass Publishers. Csíkszentmihályi, M., Abuhamdeh, S., & Nakamura, J. (2005). Flow. In Handbook of Competence and Motivation (pp. 598–608). New York, NY, USA: Guilford Press. Csíkszentmihályi, M., & Nakamura, J. (2002). The concept of flow. In The Handbook of Positive Psychology (pp. 89–92). Oxford: Oxford University Press. Csíkszentmihályi, M., & Rathunde, K. (1992). The measurement of flow in everyday life: toward a theory of emergent motivation. In Nebraska symposium on motivation (Vol. 40, pp. 57–97). Lincoln, NE, USA: University of Nebraska Press. Retrieved from http://psycnet.apa.org/psycinfo/1993-98639-002 DeLeire, T. (2000). The Wage and Employment Effects of the Americans with Disa bilities Act, 35, No. 4, 693–715. Retrieved from http://www.jstor.org/pss/146368 References 183 Deterding, S., Sicart, M., Nacke, L., O’Hara, K., & Dixon, D. (2011). Gamification. using game-design elements in non-gaming contexts. In Proceedings of the 2011 annual conference extended abstracts on Human factors in computing systems (Vol. 2, pp. 2425–2428). New York, NY, USA: ACM. doi:10.1145/1979482.1979575 Deuse, J. (2010). MTM – die Prozesssprache für ein modernes Industrial Engineering. In B. Britzke (Ed.), MTM in einer globalisierten Wirtschaft: Arbeitspro zesse systematisch gestalten und optimieren (pp. 65–79). München: Finanz buch Verlag. Dieterich, M. (1983). Geschicklichkeitserprobung mit MTM: e. Beitr. zur Diagnostik u. Förderung behinderter Mitarb. in Werkstätten für Behinderte u. Indust riebetrieben. Esslingen: Verlag Rehabilitations-Technikum. DIN EN 13199-1:2000-10. (2000). Verpackung - Kleinladungsträgersysteme - Teil 1: Allgemeine Anforderungen und Prüfverfahren [Packaging - Small Load Carrier Systems - Part 1: Common requirements and test methods]. Beuth. Retrieved from http://www.beuth.de/cmd%3Bjsessionid=RN08MOT FMB2PBOGEEGVJMFOO.4?workflowname=infoInstantdownload&doc name=9022824&contextid=beuth&servicerefname=beuth&ixos=toc Dömők, T., Szűcs, V., László, E., & Sík Lányi, C. (2012). “Break the bricks” serious game for stroke patients. In Proceedings of the 13th international confer ence on Computers Helping People with Special Needs - Volume Part I (pp. 673–680). Berlin, Heidelberg: Springer-Verlag. doi:10.1007/978-3-642 31522-0_101 Dul, J., Bruder, R., Buckle, P., Carayon, P., Falzon, P., Marras, W. S., … van der Doelen, B. (2012). A strategy for human factors/ergonomics: developing the discipline and profession. Ergonomics, 55(4), 377–395. doi:10.1080/00140139.2012.661087 Feiner, S., Macintyre, B., & Seligmann, D. (1993). Knowledge-based Augmented Re ality. Commun. ACM, 36(7), 53–62. doi:10.1145/159544.159587 Finley, K. (2012, November 14). How “Gamification” Can Make Your Customer Ser vice Worse | Wired Enterprise | Wired.com. Wired Enterprise. Retrieved November 5, 2013, from http://www.wired.com/wiredenter prise/2012/11/gamification-customer-service/ Ford, M. (2013). Could artificial intelligence create an unemployment crisis? Commun. ACM, 56(7), 37–39. doi:10.1145/2483852.2483865 Funk, M., Korn, O., & Schmidt, A. (2014a). An Augmented Workplace for Enabling User-Defined Tangibles. In CHI ’14 Proceedings of the ACM SIGCHI Con ference on Human Factors in Computing Systems. New York, NY, USA: ACM. Funk, M., Korn, O., & Schmidt, A. (2014b). Assisitive Augmentation at the Manual Assembly Workplace using In-Situ Projection. In CHI ’14 Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems. New York, NY, USA: ACM. 184 References Gee, J. P. (2007). What video games have to teach us about learning and literacy (Rev. and updated ed.). New York, NY, USA: Palgrave Macmillan. Geller, T. (2014). How Do You Feel?: Your Computer Knows. Commun. ACM, 57(1), 24–26. doi:10.1145/2555809 Gerling, K. M., Schulte, F. P., & Masuch, M. (2011). Designing and evaluating digital games for frail elderly persons. In Proceedings of the 8th International Con ference on Advances in Computer Entertainment Technology (pp. 62:1– 62:8). New York, NY, USA: ACM. doi:10.1145/2071423.2071501 Hardy, J., & Alexander, J. (2012). Toolkit support for interactive projected displays. In Proceedings of the 11th International Conference on Mobile and Ubiq uitous Multimedia (pp. 42:1–42:10). New York, NY, USA: ACM. doi:10.1145/2406367.2406419 Hassenzahl, M., Burmester, M., & Koller, F. (2008). Der User Experience (UX) auf der Spur. Zum Einsatz von www.attrakdiff.de [Tracing User Experience (UX). On using www.attrakdiff.de]. In H. Brau, S. Diefenbach, M. Hassen zahl, F. Koller, M. Peissner, & K. Röse (Eds.), Usability Professionals 2008 (pp. 78–82). Stuttgart: Fraunhofer IRB Verlag. Retrieved from http://attrak diff.de/files/up08_ux_auf_der_spur.pdf Heinz, E., Kunze, K., Gruber, M., Bannach, D., & Lukowicz, P. (2006). Using Wear able Sensors for Real-Time Recognition Tasks in Games of Martial Arts An Initial Experiment. In IEEE Symposium on Computational Intelligence and Games (pp. 98–102). IEEE. doi:10.1109/CIG.2006.311687 Huber, J., Steimle, J., Liao, C., Liu, Q., & Mühlhäuser, M. (2012). LightBeam: inter acting with augmented real-world objects in pico projections. In Proceed ings of the 11th International Conference on Mobile and Ubiquitous Multimedia (pp. 16:1–16:10). New York, NY, USA: ACM. doi:10.1145/2406367.2406388 Ikeda, N., Murray, C. J. L., & Salomon, J. A. (2009). Tracking population health based on self-reported impairments: Trends in the prevalence of hearing loss in US adults, 1976-2006. American Journal of Epidemiology, 170(1), 80–87. doi:10.1093/aje/kwp097 ISO/TC 159/SC 4. (2006). Ergonomics of human-system interaction. International Or ganization for Standardization. ISO/TC 159/SC 4. (2009). Ergonomics of human-system interaction - Part 920: Guid ance on tactile and haptic interactions. International Organization for Standardization. Kahneman, D., Krueger, A. B., Schkade, D., Schwarz, N., & Stone, A. (2004). Toward national well-being accounts. The American Economic Review, 94(2), 429– 434. Retrieved from http://www.jstor.org/stable/10.2307/3592923 Kane, S. K., Avrahami, D., Wobbrock, J. O., Harrison, B., Rea, A. D., Philipose, M., & LaMarca, A. (2009). Bonfire: a nomadic system for hybrid laptop-tab letop interaction. In Proceedings of the 22nd annual ACM symposium on User interface software and technology (pp. 129–138). New York, NY, USA: ACM. doi:10.1145/1622176.1622202 References 185 Kato, P. M., Cole, S. W., Bradlyn, A. S., & Pollock, B. H. (2008). A Video Game Improves Behavioral Outcomes in Adolescents and Young Adults With Cancer: A Randomized Trial. Pediatrics, 122(2), e305–e317. doi:10.1542/peds.2007-3134 Kern, D., & Schmidt, A. (2009). Design Space for Driver-based Automotive User Interfaces. In Proceedings of the 1st International Conference on Automo tive User Interfaces and Interactive Vehicular Applications (pp. 3–10). New York, NY, USA: ACM. doi:10.1145/1620509.1620511 Kieras, D. (1993). Using the keystroke-level model to estimate execution times. Uni versity of Michigan. Retrieved from ftp://ai.eecs.umich.edu/peo ple/kieras/GOMS/KLM.pdf Kluge, S. (2011, November 21). Methodik zur fähigkeitsbasierten Planung modularer Montagesysteme [Methodology for capability-based planning of modular assembly systems]. University of Stuttgart. Retrieved from http://elib.uni stuttgart.de/opus/volltexte/2011/6834/ Knabe, A., Rätzel, S., Schöb, R., & Weimann, J. (2010). Dissatisfied with Life but Having a Good Day: Time-use and Well-being of the Unemployed*. The Economic Journal, 120(547), 867–889. doi:10.1111/j.1468 0297.2009.02347.x Knoop, S., Vacek, S., & Dillmann, R. (2006). Sensor fusion for 3D human body track ing with an articulated 3D body model. In ICRA ’06 Proceedings of the 2006 IEEE International Conference on Robotics and Automation (pp. 1686 –1691). doi:10.1109/ROBOT.2006.1641949 Korn, O. (2012). Industrial playgrounds: how gamification helps to enrich work for elderly or impaired persons in production. In EICS ’12 Proceedings of the 4th ACM SIGCHI symposium on Engineering interactive computing sys tems (pp. 313–316). New York, NY, USA: ACM. doi:10.1145/2305484.2305539 Korn, O., Brach, M., Schmidt, A., Hörz, T., & Konrad, R. (2012). Context-Sensitive User-Centered Scalability: An Introduction Focusing on Exergames and Assistive Systems in Work Contexts. In S. Göbel, W. Müller, B. Urban, & J. Wiemeyer (Eds.), E-Learning and Games for Training, Education, Health and Sports (Vol. 7516, pp. 164–176). Berlin, Heidelberg: Springer Berlin Heidelberg. Retrieved from http://www.springerlink.com/in dex/10.1007/978-3-642-33466-5_19 Korn, O., Funk, M., Abele, S., Hörz, T., & Schmidt, A. (2014). Context-aware Assis tive Systems at the Workplace. Analyzing the Effects of Projection and Gamification. In PETRA ’14 Proceedings of the 7th International Confer ence on PErvasive Technologies Related to Assistive Environments. Korn, O., Schmidt, A., & Hörz, T. (2013). The Potentials of In-Situ-Projection for Augmented Workplaces in Production. A Study with Impaired Persons. In CHI ’13 Proceedings of the ACM SIGCHI Conference on Human Factors in Computing Systems (pp. 979–984). New York, NY, USA: ACM. doi:10.1145/2468356.2468531 186 References Kregel, J., & Dean, D. H. (2002). Sheltered vs. Supported Employment: A Direct Comparison of Long-Term Earnings Outcomes for Individuals with Cogni tive Disabilities. In J. Kregel, D. H. Dean, & P. Wehman (Eds.), Achieve ments and Challenges in Employment Services for People with Disabilities: The Longitudinal Impact of Workplace Supports Monograph. Retrieved from http://www.szentlazar.hu/media/pdf/2010/04/sheltered-vs-supported article.pdf Kronberg, A. (2013). Zwischen Pädagogik und Produktion. Qualitätsmanagement systeme in Werkstätten für behinderte Menschen [Between and Production. Quality Management in Sheltered Work Organizations]. Lützelsdorf, Ger many: Rossol. Retrieved from http://www.verlag-rossol.de/titel/kronberg qm-in-wfbm/ Kühn, W. (2006). Digital factory: simulation enhancing the product and production engineering process. In Proceedings of the 38th conference on Winter sim ulation (pp. 1899–1906). Retrieved from http://dl.acm.org/cita tion.cfm?id=1218458 Langheinrich, M., Schmidt, A., Davies, N., & José, R. (2013). A Practical Framework for Ethics: The PD-net Approach to Supporting Ethics Compliance in Pub lic Display Studies. In Proceedings of the 2Nd ACM International Sympo sium on Pervasive Displays (pp. 139–143). New York, NY, USA: ACM. doi:10.1145/2491568.2491598 Lazar, J., & Hochheiser, H. (2013). Legal Aspects of Interface Accessibility in the U.S. Commun. ACM, 56(12), 74–80. doi:10.1145/2500498 Lee, E. A. (2008). Cyber physical systems: Design challenges. In 11th IEEE Interna tional Symposium on Object Oriented Real-Time Distributed Computing (ISORC) (pp. 363–369). Retrieved from http://ieeex plore.ieee.org/xpls/abs_all.jsp?arnumber=4519604 Letessier, J., & Bérard, F. (2004). Visual tracking of bare fingers for interactive sur faces. In Proceedings of the 17th annual ACM symposium on User interface software and technology (pp. 119–122). New York, NY, USA: ACM. doi:10.1145/1029632.1029652 Lewis, J. R., & Sauro, J. (2009). The Factor Structure of the System Usability Scale. In Proceedings of the 1st International Conference on Human Centered De sign: Held as Part of HCI International 2009 (pp. 94–103). Berlin, Heidel berg: Springer-Verlag. doi:10.1007/978-3-642-02806-9_12 Lopez, A. D., Mathers, C. D., Ezzati, M., Jamison, D. T., & Murray, C. J. L. (2006). Measuring the Global Burden of Disease and Risk Factors, 1990–2001. In A. D. Lopez, C. D. Mathers, M. Ezzati, D. T. Jamison, & C. J. Murray (Eds.), Global Burden of Disease and Risk Factors. Washington (DC), USA: World Bank. Retrieved from http://www.ncbi.nlm.nih.gov/books/NBK11817/ Luger, G. F. (2009). Artificial intelligence: structures and strategies for complex problem solving (6th ed.). Boston: Pearson Addison-Wesley. References 187 Makedon, F., Le, Z., Huang, H., Becker, E., & Kosmopoulos, D. (2009). An Event Driven Framework for Assistive CPS Environments. SIGBED Rev., 6(2), 3:1–3:9. doi:10.1145/1859823.1859826 Manzeschke, A., Weber, K., Rother, E., & Fangerau, H. (2013). Ethische Fragen im Bereich altersgerechter Assistenzsysteme: Ergebnisse der Studie [Ethical questions in the area of assistive systems for the elderly - results of a study]. Berlin: VDI/VDE Innovation + Technik. Maynard, H. B., Stegemerten, G. J., & Schwab, J. L. (1948). Methods-time measure ment. New York, NY, USA: McGraw-Hill. McGonigal, J. (2011). Reality is broken: Why games make us better and how they can change the world. Penguin books. Mink, J. A. (1975). MTM and the disabled. The MTM Journal of Methods Time Meas urement, 2(2), 23–28. Mitcham, C. (1994). Thinking through technology: the path between engineering and philosophy. Chicago: University of Chicago Press. Morelli, T., Foley, J., & Folmer, E. (2010). Vi-bowling: a tactile spatial exergame for individuals with visual impairments. In Proceedings of the 12th interna tional ACM SIGACCESS conference on Computers and accessibility (pp. 179–186). New York, NY, USA: ACM. doi:10.1145/1878803.1878836 Murphy, S. T., & Rogan, P. M. (1995). Closing the shop: conversion from sheltered to integrated work. Paul H. Brookes Publishing Company. Nevins, A. (1954). Ford: The Times, the Man, the Company. New York, NY, USA: Scribner. Nielsen, J. (1993). Usability Engineering. San Francisco, CA, USA: Morgan Kauf mann Publishers Inc. Nilsson, S., & Johansson, B. (2008). Acceptance of augmented reality instructions in a real work setting. In CHI ’08 Extended Abstracts on Human Factors in Computing Systems (pp. 2025–2032). New York, NY, USA: ACM. doi:10.1145/1358628.1358633 Nunes, F., Silva, P. A., & Abrantes, F. (2010). Human-computer interaction and the older adult. An Example Using User Research and Personas. In PETRA ’10 Proceedings of the 3rd International Conference on PErvasive Technolo gies Related to Assistive Environments (p. 1). ACM Press. doi:10.1145/1839294.1839353 Pagulayan, R. J., Keeker, K., Wixon, D., Romero, R. L., & Fuller, T. (2012). UserCentered Design in Games. In J. A. Jacko & A. Sears (Eds.), The humancomputer interaction handbook: fundamentals, evolving technologies, and emerging applications (pp. 795–821). Boca Raton, FL, USA: CRC Press. Retrieved from http://dl.acm.org/citation.cfm?id=772072.772128 Pinhanez, C. S. (2001). The Everywhere Displays Projector: A Device to Create Ubiq uitous Graphical Interfaces. In Proceedings of the 3rd international confer ence on Ubiquitous Computing (pp. 315–331). London, UK: SpringerVerlag. Retrieved from http://dl.acm.org/citation.cfm?id=647987.741324 188 References Polanyi, K. (1957). Aristotle Discovers the Economy. In K. Polanyi, C. M. Arsenberg, & H. W. Pearson (Eds.), Trade and market in the early empires: economies in history and theory. New York, NY, USA: Free Press. Prensky, M. (2007). Digital game-based learning. St. Paul, Minn., USA: Paragon House. Putnam, C., & Cheng, J. (2013). Helping therapists make evidence-based decisions about commercial motion gaming. SIGACCESS Access. Comput., (107), 3– 10. doi:10.1145/2535803.2535804 Quick, J. H., Duncan, J. H., & Malcom, J. A. (1962). Work-Factor Time Standards. Measurement of manual and mental work. New York, NY, USA: McGrawHill Book Co. Reeves, B., & Read, J. L. (2009). Total Engagement: Using Games and Virtual Worlds to Change the Way People Work and Businesses Compete. Harvard Business Press. Robine, J.-M., & Michel, J.-P. (2004). Looking forward to a general theory on popu lation aging. The Journals of Gerontology. Series A, Biological Sciences and Medical Sciences, 59(6), M590–597. Rogers, Y., & Marsden, G. (2013). Does he take sugar? Moving beyond the rhetoric of compassion. Interactions, 20(4), 48–57. Retrieved from http://dl.acm.org/citation.cfm?id=2486238 Runeson, P., & Höst, M. (2008). Guidelines for conducting and reporting case study research in software engineering. Empirical Software Engineering, 14(2), 131–164. doi:10.1007/s10664-008-9102-8 Rüther, S., Hermann, T., Mracek, M., Kopp, S., & Steil, J. (2013). An assistance sys tem for guiding workers in central sterilization supply departments. In Pro ceedings of the 6th International Conference on PErvasive Technologies Related to Assistive Environments (pp. 3:1–3:8). New York, NY, USA: ACM. doi:10.1145/2504335.2504338 Salthouse, T. A. (1990). Working memory as a processing resource in cognitive aging. Developmental Review, 10(1), 101–124. doi:10.1016/0273-2297(90)90006 P Sardis, E., Voulodimos, A., Anagnostopoulos, V., Lalos, C., Doulamis, A., & Kos mopoulos, D. (2010). An industrial video surveillance system for quality assurance of a manufactory assembly. In PETRA ’10 Proceedings of the 3rd International Conference on PErvasive Technologies Related to Assistive Environments (p. 66). Satre, D. D., Knight, B. G., & David, S. (2006). Cognitive-behavioral interventions with older adults: Integrating clinical and gerontological research. Professi onal Psychology: Research and Practice, 37(5), 489–498. doi:10.1037/0735-7028.37.5.489 Schlaffer, H. (2005). Poesie und Wissen: die Entstehung des ästhetischen Bewusst seins und der philologischen Erkenntnis. Frankfurt am Main: Suhrkamp. References 189 Schmidt, A. (2000). Implicit human computer interaction through context. Personal Technologies, 4(2-3), 191–199. Retrieved from http://link.springer.com/ar ticle/10.1007/BF01324126 Schmidt, A., Kranz, M., & Holleis, P. (2005). Interacting with the Ubiquitous Com puter: Towards Embedding Interaction. In Proceedings of the 2005 Joint Conference on Smart Objects and Ambient Intelligence: Innovative Context-aware Services: Usages and Technologies (pp. 147–152). New York, NY, USA: ACM. doi:10.1145/1107548.1107588 SGB IX. Sozialgesetzbuch (SGB) Neuntes Buch (IX) – Rehabilitation und Teilhabe behinderter Menschen – (SGB IX) (2001). Shneiderman, B. (2010). Designing the user interface: strategies for effective humancomputer interaction (5th ed.). Boston: Addison-Wesley. Shotton, J., Fitzgibbon, A., Cook, M., Sharp, T., Finocchio, M., Moore, R., … Blake, A. (2011). Real-time human pose recognition in parts from single depth im ages. In Proceedings of the 24th IEEE Conference on Computer Vision and Pattern Recognition (Vol. 2, p. 7). Siddiqui, M., & Medioni, G. (2010). Human pose estimation from a single view point, real-time range sensor. In CVPRW ’10 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops 2010 (pp. 1 –8). doi:10.1109/CVPRW.2010.5543618 Suits, B. (2005). The grasshopper: games, life and utopia. Peterborough, Ont., USA: Broadview Press. Sykes, J., & Brown, S. (2003). Affective gaming: measuring emotion through the gamepad. In CHI’03 extended abstracts on Human factors in computing systems (pp. 732–733). Retrieved from http://dl.acm.org/cita tion.cfm?id=765957 Ten, S. (2010). How Kinect depth sensor works - stereo triangulation? Mirror Image. Retrieved November 28, 2013, from http://mirror2image.word press.com/2010/11/30/how-kinect-works-stereo-triangulation/ United Nations. Convention on the Rights of Persons with Disabilities (2008). Re trieved from http://hpod.pmhclients.com/pdf/ConventionImplications.pdf United Nations, Department of Economic and Social Affairs, Population Division. (2013). World Population Prospects: The 2012 Revision (No. Volume I: Comprehensive Tables). Vanderheiden, G. C. (2008). Ubiquitous Accessibility, Common Technology Core, and Micro Assistive Technology: Commentary on “Computers and People with Disabilities” ACM Trans. Access. Comput., 1(2), 10:1– 10:7. doi:10.1145/1408760.1408764 VDI Verein Deutscher Ingenieure. (1990). VDI 2860 Montage- und Handhabung stechnik [Technology for Assembly and Handling]. Beuth. Retrieved from http://www.vdi.de/uploads/tx_vdirili/pdf/2372581.pdf VDI Verein Deutscher Ingenieure. (2008). VDI 4499 Grundlagen der Digitalen Fab rik. Digital factory Fundamentals. Beuth. Retrieved from http://www.vdi.de/uploads/tx_vdirili/pdf/9856297.pdf 190 References Von Ahn, L., & Dabbish, L. (2008). Designing Games with a Purpose. Commun. ACM, 51(8), 58–67. doi:10.1145/1378704.1378719 Weaver, K. A., Baumann, H., Starner, T., Iben, H., & Lawo, M. (2010). An empirical task analysis of warehouse order picking using head-mounted displays. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (pp. 1695–1704). New York, NY, USA: ACM. doi:10.1145/1753326.1753580 Weber, M. (2009). The Protestant ethic and the spirit of capitalism: the Talcott Par sons translation interpretations (1st ed.). New York: W.W. Norton & Co. Weiser, M. (1999). The computer for the 21st century. SIGMOBILE Mob. Comput. Commun. Rev., 3(3), 3–11. doi:10.1145/329124.329126 Wellner, P. (1991). The DigitalDesk calculator: tangible manipulation on a desk top display. In Proceedings of the 4th annual ACM symposium on User inter face software and technology (pp. 27–33). New York, NY, USA: ACM. doi:10.1145/120782.120785 Westkämper, E. (2006). Digitale Produktionm [Digital Production]. In H.-J. Bullinger (Ed.), Technologieführer: Grundlagen - Anwendungen - Trends [Techno logy Guide: Principles - Applications - Trends] (pp. 435–439). Heidelberg, Germany: Springer. Wilson, A. D. (2010). Using a Depth Camera As a Touch Sensor. In ACM Interna tional Conference on Interactive Tabletops and Surfaces (pp. 69–72). New York, NY, USA: ACM. doi:10.1145/1936652.1936665 World Health Organisation. (2001). The International Classification of Functioning, Disability and Health (ICF). Retrieved October 29, 2013, from http://www.who.int/classifications/icf/en/ World Health Organisation, & World Bank. (2011). World report on disability. Ge neva; Washington, DC: World Health Organization ; World Bank. World Health Organization. (2004). World Health Survey. Retrieved from http://www.who.int/healthinfo/survey/en/ Zäh, M. F., Wiesbeck, M., Engstler, F., Friesdorf, F., Schubö, A., Stork, S., … Wall hoff, F. (2007). Kognitive Assistenzsysteme in der manuellen Montage Adaptive Montageführung mittels zustandsbasierter, umgebungsabhängi ger Anweisungsgenerierung [Cognitive assistance in manual assembly – Adaptive assembly guidance through a context-sensitive generation of in structions]. wt-online, (9), 644–650. Zhou, J., Lee, I., Thomas, B., Menassa, R., Farrant, A., & Sansome, A. (2011). Ap plying spatial augmented reality to facilitate in-situ support for automotive spot welding inspection. In Proceedings of the 10th International Confer ence on Virtual Reality Continuum and Its Applications in Industry (pp. 195–200). New York, NY, USA: ACM. doi:10.1145/2087756.2087784 191 Index 11 INDEX AAL 16 Ethical standards 4, 111, 162 ACM 36 EU 16 ActionScript 107 Exergames 55 Adjoined spheres approach 90 Affective computing 164 Facial Action Coding System (FACS) 86 AIR 107 Flow 30, 104 Ambient assisted living 16 Flow curve 82 Applied games 53 Games engineering 79, 103 Artificial intelligence 2, 38, 53, 161 Games with a purpose 53 ASED 93, 96, 100, 107 Gamification 32, 53, 82, 85, 103, 147 ASLM 8 HCI 33, 36, 67 ASSETS 36 HMI 44, 57, 69 Augmented reality 37, 51, 60 ICF 1, 16 Bacon, Francis 24 IEEE 36, 51 Burnout 25 Impaired persons 16, 21, 26, 28, 35, 53, 71, 103, 148, 161 BWH 110 CAD 44, 57 CNC 44, 69 Companions 39 Cyber-physical systems 37, 46, 161 Demographic change 4, 15, 161 Digital factory 44 Digital natives 53 Disabled persons 16, 26 Elderly persons 14, 20, 35, 53, 71, 161 ERP 45 Implicit interaction 33, 69 Inclusion 22, 26, 161 ISO 9241 58, 67 Java 93, 102 Kinect 34, 48, 89, 102 Leap Motion 163 Lee, Edward 46 Luther, Martin 24 McGonigal, Jane 103 MEESTAR 14, 65 Minsky, Marvin 27, 38 192 Index Mixed reality 37 Structured light 49, 50, 89 MTM 33, 40, 47 Supported work 21 Natural interaction 34, 48, 51, 70 Tetris 104 Nielsen, Jakob 68 Time-of-flight 49, 89, 163 NITE 49, 167 ToF 49, 89, 163 Old-age support ratio 16, 21 Total quality 57, 70 OpenCV 93 Turing, Alan 38 OpenNI 49, 167 Ubiquitous computing 33, 36 PETRA 36 PHP 120 UN Convention on the rights of persons with disabilities 26, 162 Playful design 53 Universal access 36 PLC 58, 60 Universal accessibility 71 PLM 45 Universal design 36, 71 Poka-Yoke 57 UX 12, 27, 39, 63 Polanyi, Karl 25 Virtual factory 36, 45 PPS 44 Virtual reality 37 Prensky, Marc 53 Weber, Max 24 PrimeSense 49 Weiser, Mark 33, 51 Project Natal 34 Westkämper, Engelbert 45 Quantified self 162 WHO 16, 20 REFA 40, 41 Wii 34 RFID 46 Wizards 1, 39 Serious games 53 Work-life balance 25 Sheltered work 3, 21 Xbox 55, 163 Shneiderman, Ben 67 XBox 34 Smart watch 85 Xtion 48 Socrates 23 Z-values 89, 155 While context-aware assistive systems (CAAS) have become ubiquitous in cars or smartphones, most workers in production environments still rely on their skills and expertise to make the right choices and movements. The quality gates currently established in the industry successfully remove failed work results; however, they usually operate in a spatial and temporal distance from the workplace and the worker. Thus workers lack the opportunity to learn from problems on the fly and to improve their work habits. Today’s motion recognition systems and micro projectors allow to display work-relevant information directly into a worker’s field of vision. Thus the technical pre-requisites for CAAS with instant feedback at the workplace (in situ) essentially have been established. Although every worker would benefit from CAAS, persons with impairments and elderly persons with reduced physical and mental capabilities require such systems the most. CAAS have the potential to empower these persons to do more complex work or to remain in regular jobs longer. Thus they combine economic benefits with inclusion and address the demographic change. After an overview of the relevant backgrounds from ethics, psychology, computer science and engineering as well as the relevant state-of-the-art, we establish requirements which result in a model for ideal CAAS. As the framework aims to improve not only the work results but also the workers’ motivation, the model incorporates elements from game design. An exemplary implementation covering essential aspects of the model is described. The effects of both the augmentation by projection and by gamification are evaluated in a study with impaired persons. An additional focus lies on the ethical implications of assistive systems which supervise and model their users in real-time. About the author: Oliver Korn is a computer scientist and CEO of the software company Korion. This is his PhD thesis on interactive assistive systems and gamification. It was realized at the Institute for Visualization and Interactive Systems (VIS) of the University of Stuttgart. He studied Computational Linguistics as well as English and German Language at the universities of Stuttgart and Glasgow. Since 2001 he works in projects focusing on human computer interaction, assistive systems, gaming and simulations. 2003 he co-founded the Fraunhofer spin-off Korion. He worked as associate lecturer for the Karlsruhe Institute for Technology (KIT) and lectures at the Steinbeis Academy, the Stuttgart Media University (HdM) and the University of Stuttgart. He also is a certified Project Manager (IHK) and Professional Member of the Association for Computing Machinery (ACM). EUR 59,95 ISBN 9781291864861 90000 9 781291 864861