supporting mobile microinteractions
Transcription
supporting mobile microinteractions
SUPPORTING MOBILE MICROINTERACTIONS A Thesis Proposal Presented to The Academic Faculty by Daniel L. Ashbrook In Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the College of Computing Georgia Institute of Technology December 2007 SUPPORTING MOBILE MICROINTERACTIONS Approved by: Thad E. Starner, Advisor College of Computing Georgia Institute of Technology Blair MacIntyre College of Computing Georgia Institute of Technology Gregory D. Abowd College of Computing Georgia Institute of Technology James A. Landay Computer Science & Engineering University of Washington Charles L. Isbell, Jr. College of Computing Georgia Institute of Technology Date Approved: TABLE OF CONTENTS LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.1 Thesis Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.2 Two Interaction Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2.1 Touchscreen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.2.2 Gestural . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.3 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.4 Proposal Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 RELATED WORK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2.1 Mobility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2.2 Touchscreen Watch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 2.3 Gesture Watch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 MICROINTERACTIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 3.1 Calendar Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 3.2 Quickdraw . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 3.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 3.2.2 Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 3.2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 3.2.4 Quickdraw Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 TOUCHWATCH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 4.1 Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 4.1.1 Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 4.1.2 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 4.1.3 Apparatus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 4.1.4 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 4.1.5 Participants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 I II III 3.3 IV iii 4.1.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 4.1.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Rim Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 4.2.1 Card Dragging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 4.2.2 Calendar Dialing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 Proposed Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 WRIST-BASED GESTURE INTERACTION . . . . . . . . . . . . . . . . . . . . . . 29 5.1 Gesture Triggering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 5.2 Proposed work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 5.2.1 Hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 5.2.2 Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 5.2.3 Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 PLAN FOR COMPLETION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 4.2 4.3 V VI iv LIST OF TABLES 1 In seconds, average time for events during scheduling appointments, per commonly used device. Navigation time for scrap paper is effectively zero. . . . . . . . . . . 11 Claimed (columns) versus actual (rows) usage of devices for scheduling task. The rows are sorted by retrieval + navigation time, from Table 1, shown in parentheses. 12 Devices carried by participants and their location. “Body” refers to on-body placement of a device, such as clipping to clothing or using a wrist strap. “Bag” includes purses and backpacks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 4 Statistics for mean (SD) hand, pocket and access times in seconds. . . . . . . . . . 17 5 Mean (SD) effective width We in pixels for each movement condition, parallel (k) and perpendicular (⊥) to the direction of movement. . . . . . . . . . . . . . . . . 23 2 3 v LIST OF FIGURES 1 The Tissot T-Touch (a) and Silen-T (b) touch-sensitive watches. The Silen-T’s touch mechanism is activated by touching the crown (next to the E on the bezel) for longer than one second; the areas on the crystal labeled THERMO, METEO, ALTIMETER , CHRONO, COMPASS , and ALARM then become sensitive, activating various functionality when touched. The Silen-T also features a touch-sensitive crystal, which is activated by pressing the crown. The watch then vibrates when the user’s finger is run clockwise around the face, with a long vibration when the finger is at the current hour position and short vibrations at the minute position. . . . . . . . . . . . . . 7 The IBM WatchPad (a) and Blaskó’s interface for a variation with nine touchsensitive areas. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 The Fossil/Abacus WristPDA. (a) shows the application menu, with icons large enough to push with a fingertip; (b) shows the calculator application with tiny buttons; (c) illustrates the scale of the included stylus that is normally tucked into the watch band. It is likely that the stylus or another pointed instrument would be needed to effectively use the calculator application. . . . . . . . . . . . . . . . . . 9 Interfaces using a center/rim style of interaction. (a) is an example of a pie menu [5], (d) illustrates Cirrin [17], (b) is a conceptual visualization of the earPod menu system [38], and (c) is a pie menu-like selection system designed for see-through head-mounted displays (the central black area will appear to be transparent to the wearer). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Three on-body command-gesture devices. Device (a) is the Gesture Pendant [1], (b) is an example of FreeDigiter [20] in use, and (c) illustrates the Gesture Watch [12]. 10 The path participants walked, starting at flag 1 and proceeding either clockwise or counterclockwise. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 7 Placement: pocket (a), hip (b), and wrist (c). . . . . . . . . . . . . . . . . . . . . 14 8 Screen (a) is shown when an alert occurs. The participant must mentally note the displayed number and slide the blue box from the left to the target box on the right. Screen (b) is shown after the slide is completed; from the displayed four numbers, the participant must touch the number that was displayed on screen (a). . . . . . . 15 Timeline of events and status of brightness detector during one notification-response cycle. The top line is the percentage of light detected from the camera (complete darkness, 0% on the bottom), and the bottom is the timeline of events recorded by the phone’s logging software. See the text for further details. . . . . . . . . . . . . 16 2 3 4 5 6 9 vi 10 Images (a), (b) and (c) show the targets (short green lines at rim of circle) and movement guides (yellow line between targets) displayed to study participants. Images (d), (e) and (f) show actual movements for one session; blue lines indicate movement by participant’s finger and small yellow circles show inflection points. Movement images were displayed only to researcher. Images (a) and (d) are for tap condition; images (b) and (e) are for through condition; images (c) and (f) are for rim condition. Images (a)–(f) are printed at actual size. . . . . . . . . . . . . . . 21 11 The simulated touchscreen watch. . . . . . . . . . . . . . . . . . . . . . . . . . . 22 12 Effective index of difficulty (IDe ) versus movement time (MT ) for the three movement conditions. The regression line is dotted in each graph. Regression statistics are: tap slope: 40ms/bit, intercept: 228ms, R2 : 0.04); through slope: 54ms/bit, intercept: 106ms, R2 : 0.09); rim slope: 158ms/bit, intercept: −18ms, R2 : 0.45). . 24 13 Possible watch face layouts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 14 Blaskó’s card dragging technique [9]. The default watch face is illustrated in (a), with the asterisk denoting the desired card. The motion made by the user’s finger is illustrated in (b), and the resulting screen with the activated content is shown in (c). 25 Proposed “card dragging” interaction using icons on rim of watch. (a): view of watch with no applications active; (b): user touches face to show application icons; (c): user selects desired application icon; (d)–(f): user drags application “card” out of “stack”. Images are shown at actual size for hardware used during Fitts’ study and conform to design recommendations discovered therein. . . . . . . . . . 26 Mockup of a possible dial-metaphor interface for calendar navigation. Dark gray segments in (a)–(c) show “now”, thick red borders indicate area selected, and bands around rim in (d) indicate appointments. User touches red-banded area and drags finger clockwise to zoom in to next closer detail view; to zoom out, user drags finger counterclockwise. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 Candidate gesture performed by user (a) and partial gesture database (b) with match percentage (bottom) and database contents (top) illustrated. . . . . . . . . 30 Prototype watch hardware. Features on the side of the board illustrated in (a), from the top left and in clockwise direction are the Bluetooth module, the power/charge switch, battery wires, reset and action buttons, display connector, and mini-USB connector (for charging and programming). (b) shows (in the same order), the microSD card slot, unpopulated pads for the accelerometer chip, and the processor. Also visible are programmable LEDs and Bluetooth status LEDs. . . . . . . . . . 31 15 16 17 18 vii ABSTRACT Many interactions with mobile devices could be more like checking the time on a wristwatch—short and almost without conscious thought. I term activities like these microinteractions. For my dissertation work, I propose to investigate and design interfaces to facilitate microinteractions. I will do this through developing interfaces for small, on-body devices. Many interactions that might otherwise be micro- are stymied by too-long “setup” times—the amount of time it takes to retrieve the device and navigate to the appropriate application—that must take place before the desired interaction can be accomplished. My proposed work will shorten or remove altogether the setup time The lack of support for microinteractions in current technology leads to a failure to use the provided functionality, to the detriment of the user. To better enable microinteractions, I propose to develop two new techniques for wrist-based interactions. I choose the wrist because it is a socially acceptable platform for wearable devices, and because it enables very fast access. The two interactions are touchscreen-based and gesture-based. One issue with mobile interactions is how to avoid false activations, when the device is unintentionally activated by some natural action of the user. Many current devices attempt to solve the problem by having “push-to-activate” mechanisms; however, this practice results in added complexity and, in some cases, the loss of single-handed functionality. The methods that I explore in this proposal remove the need for push-to-activate functionality, allowing the user to initiate microinteractions with no prelude. For the touchscreen-based interactions, I propose to use a round-faced touchscreen watch. Historically, most clocks and watches have been circular, while all commercially available touchscreen watches thus far have been rectangular. I present results of a study on the best method for placing buttons around the rim of a round touchscreen watch, and I propose two related interaction techniques that simultaneously encourage users to utilize the most efficient method of interaction and prevent undesired interaction due to accidental touches of the screen. Mobile gesture-based systems in particular suffer from the issue of unintentional activation: the normal movements of the user can be confused for intentional gestures. This problem is most often solved with push-to-activate; however, in the case of wrist-based gesture systems, pushing requires the use of the non-wristwatch hand, meaning that the gesture functionality could easily be replaced by buttons. I propose a solution to the unintentional activation problem that, rather than involving pushing-to-activate, uses an “Everyday Gesture Library” to help interaction designers choose gestures that are unlikely to be confused with normal gestures made in the course of daily life. 1 CHAPTER I INTRODUCTION I propose to investigate the space of microinteractions with small, on-body devices. The category of small, on-body devices is reasonably self-descriptive. I define microinteractions as single-purpose interactions with a device that take a very short amount of time, such as reading a text message, dialing a phone number, or checking upcoming appointments. Often, would-be microinteractions are surrounded by “setup” and “teardown” interactions that serve only to prepare the device for the microinteraction. An example of this is reading a text message, or SMS, on a mobile phone. When the SMS is received, the user is alerted via a sound or vibration. To read the SMS, the user has to remove the phone from its storage place (a bag or pocket, for example), open or unlock the phone, and navigate through one or more menus to get to the message. After reading the message, a similar sequence must be followed by closing the application, re-locking or closing the phone, and returning it to the original storage location. One way of better enabling such interactions to be more “micro” is by reducing the setup and teardown times surrounding the interactions. Of these two times, the setup time is more important, as it represents the delay between the desire by the user to perform some action, and the actual time at which the user is able to start the action. The teardown time, on the other hand, usually only delays the user in returning to the “natural state” that he was in before the interaction started. I propose the use of on-body devices to reduce both setup and teardown times. As Chapter 3 reveals, removing the need to extract a device from a pocket or other storage area greatly decreases the setup time for an interaction. Making a device small and placing it in an easily accessible location on the body enables the user to quickly access the device, potentially eliminating the need to perform any teardown actions at all. As an example of an on-body device with which one might have microinteractions, I specifically propose to investigate the wristwatch. Watches are socially accepted as wearable technology, and judging by the consumer market for “gadget” watches, there is an appetite for watches that support microinteractions. For example, Sony recently released a watch/phone combo that allows the audio player functionality of the phone to be controlled by the watch. Both input and output are important when considering the space of interaction with small, onbody devices such as wristwatches. In this proposal, I focus on the problem of input. I approach this problem with two input techniques: touchscreen-based and gesture-based. 1.1 Thesis Statement My hypothesis is that mobile microinteractions can be supported with two techniques: • Rim-based interactions on a round touchscreen wristwatch will allow slower but complex interaction while avoiding false positives; and • Push-to-activate mechanisms can be avoided for fast wrist-based gestural interactions by creating and using an Everyday Gesture Library to help in the selection of appropriate gestures. 2 1.2 Two Interaction Techniques Although I am concentrating on the wristwatch as an exemplar of small on-body devices, there is the question of how to interact with the watch. Consider these potential applications: checking the time, skipping to the next track on a digital audio player, sending a call to voicemail, reading a text message, checking the arrival time of the next bus, taking down a phone number, getting an idea of appointment times for the day, dialing a phone number from an address book, calculating a tip, checking the weather, updating status information in a social notification service such as “Twitter”, and setting or responding to a reminder alarm. There are two categories these applications can be divided into: 1. User initiating: The user wants to perform an action, and can decide when to do it. Examples include dialing a phone number, updating social notification services, or reading a text message. 2. User responding: The user responds to an externally-imposed event such as an incoming call or a reminder alarm going off. The event is time-sensitive, so the user has little latitude in deciding when to respond. To complement each of these interaction styles, I propose to investigate two interaction techniques: 1.2.1 Touchscreen For user-initiated tasks such as inputting a phone number, it makes sense to assume that the user can choose somewhat when and where to initiate that task; for example, by stepping out of the flow of people on the sidewalk to briefly interact. In this case, the ability to have more complex input and output are more important than just speed of access. Touch is one well-understood way to interact with a small device; however, many touchscreen-based watches require the use of a tiny stylus due to the small screen size. In Chapter 4 I discuss the problem of using the finger to interact with a tiny touchscreen, present results on the accuracy of several types of fingertip-based interaction on a round touchscreen watch, and propose an interaction technique that allows the avoidance of accidental activation by a sleeve or brush of the hand. 1.2.2 Gestural When an event occurs to which a user must respond quickly, a gesture is a faster option than having to manually manipulate a device. To send a call to voicemail, silence an alarm, or to skip to the next audio track, the user can use a quick and simple “twitch” motion. Gesture recognition can augment or replace context sensitivity in some cases: rather than the phone deciding not to ring based on context, it can first perform a short vibration at the wrist and allow the user to silence it very quickly and unobtrusively with a short gesture, allowing the user to continue the current activity without interruption. One problem with this idea is which gestures to use. To avoid having to use the other hand to push a button for “triggering,” it is desirable for the system to perform continuous recognition. Doing so, however, raises the possibility of the system mistaking a normal gesture made in conversation for a command to the system. I propose a solution to this problem in Chapter 5. 1.3 Contributions The contributions of the proposed work are several. Supporting mobile microinteractions by reducing access time and avoiding push-to-activate will affect everyday interactions with mobile devices. By allowing more efficient usage, users will be more likely to use the functionality that the device offers, and speeding up access time for the many daily uses that devices are put to will have a large overall effect. My proposed work has several general contributions: 1. Decreasing access time by removing the “push to activate” mechanism present in many systems [1, 19, 28]. Doing this not only removes a time-consuming motion, but can reduce the overall complexity of interaction by allowing one-handed use. Chapter 4.2 discusses this with respect to round touchscreen watches, and while Chapter 5.1 concentrates on wristbased gesture interfaces, the techniques discussed are applicable to a wide range of gesture interfaces for everyday use. 2. The introduction of new interaction techniques for the wrist. The wristwatch is a piece of technology that is well-understood by the population at large, and that is starting to garner interest as a platform for interaction. Most extant methods for interacting with the wristwatch are, however, button-based, often leading to unnecessary complexity. In Chapters 4.2 and 5.1, I propose two new methods for interacting with a wristwatch that could lead to simpler or more efficient interactions with multifunctional wristwatches. 3. A study quantifying the access time properties of mobile devices stored in the pocket, in a belt holster, and mounted on the wrist. Although one would expect that a watch is faster to access than a pocket or holster, I quantify how much faster it is and break down the access time into several measurable steps. This study, presented in Chapter 3.2, also indicates that people are generally capable of simple interactions with a touchscreen while performing pedestrian navigation. My next set of contributions relates to touchscreen wristwatches with round faces: 4. A study determining the optimal on-screen button size for three interaction types on a round touchscreen watch. As discussed later, a round display suggests an interaction region around the edge and a separate center area. Given the edge-based interaction area, what type of touch interaction should be done, and how many buttons may be accommodated? This study, presented in Chapter 4.1, compares three types of touches—tapping, sliding in a straight line, and sliding along the rim—and determines the optimal button size for each. 5. A low-false-positive interaction technique. To remove the need for a “push-to-touch” mechanism while avoiding false positives, there must be a way to distinguish intentional touches of the screen from accidental impacts, such as by a sleeve or accidental brush of the hand. In Chapter 4.2 I introduce an interaction technique that avoids false detection of user actions by using a dial metaphor. 6. Experimental validation of the study results. In Chapter 4.3, I propose a study to experimentally validate the results of the button-size study. The results of the first study were found using a Fitts-style reciprocal pointing task with very small targets; thus, I plan to re-run the study to determine user performance with actual buttons of sizes that were found. The final set of contributions is related to gesture-based interactions: 7. A method for interaction designers to design low-false-positive gestures. To allow users to initiate actions using gestures while avoiding a push-to-gesture mechanism, in Chapter 5.1 I propose a system to allow interaction designers to test potential gestures against a database of everyday movements made by typical users. This system will allow designers to pick gestures that have a low probability of being falsely recognized. 8. Prototype gesture-sensitive watch hardware. Experimentation with wearable sensing is often difficult, requiring the carrying of extra hardware such as large batteries, palmtop computers and various cables. I will publish schematics and software for the gesture-sensing watch platform described in Chapter 5.2.1 to enable future progress by other researchers. 1.4 Proposal Overview First, in Chapter 2 I outline some related work in mobile interaction, small on-body devices, and research relevant to the proposed interaction techniques. Chapter 3 presents and discusses some quantitative data that help to explain microinteractions. In Chapter 4 I discuss my progress so far with touchscreen wristwatch interactions and propose further work, and in Chapter 5 I do the same for gestural interaction. Finally, Chapter 6 presents my plan for completion of my dissertation. CHAPTER II RELATED WORK This chapter presents previous research in areas related to this proposal. Some of the cited work is only generally related, as little work has been done in some of the specific areas that I am proposing to investigate. 2.1 Mobility The use of mobile technology while on the go is an area of research that has been getting more attention recently. This is unsurprising: as electronics get smaller and more power-efficient, they lend themselves to easily being carried. At the same time, the increasing prevalence of wireless services means that communication has become supported in places that were previously only “transition zones” such as sidewalks, subways and highways. These developments mean that people now have the capability to interact and communicate in places where they previously could only walk, ride or drive. For better or for worse, on-the-go device usage is an activity that will need to be taken into account when designing interaction. Kristoffersen and Ljungberg noted a variety of problems with portable devices that combined to prevent activity from simply “taking place”—instead, users of the technology had to “make place” to use the device [13]. The biggest problems they found appeared to be related to the form-factor of the devices used, which were PDAs: users needed two hands to use the device, but often only one or no hands were free; and frequently they needed to put down the computer to use it—either to have a surface to type or write on, or to refer back to the display—but no surface was available for this purpose. These results suggest that, for some purposes, a wrist-based platform may be more practical. In contrast to the physical limitations of devices, the Resource Competition Framework proposed by Oulasvirta et al. [23] provides a way to think about the cognitive demands placed on the user of a mobile device while in various situations. The authors performed an experiment, asking users to perform various mobile Internet tasks in a variety of situations. They found that visual attention paid to the environment versus the device varied from 51% (on a busy street), to 22% (in a café), to 14% (on the metro), implying that interfaces allowing eyes-free interaction may be more usable in situations involving high competition for the user’s attention. Patel et al. examined people’s perceptions about how often they have their mobile phone nearby. Their data show that people routinely overestimate the physical availability of their mobile phones [24]. Even when the mobile phone is with its user, it may not be quickly accessible; Cui et al. found that 40% of women and 30% of men miss phone calls simply due to the manner in which they carry the mobile phone on their person [8]. Similarly, Starner et al. found correlations between an individual’s decision to use or not use a mobile scheduling device (such as a day planner or PDA) and the amount of time and effort required to access and make ready the device [34]. Together, these studies suggest that the time required to access a device may be an important property affecting mobile use. 6 2.2 Touchscreen Watch Despite the increasing functionality in wrist-worn technology, there has not been much academic work on wristwatch interfaces, and I have been unable to find any work at all on round touchscreen watches, either commercially or in the literature. Two commercial examples of round-faced watches with touch capability—though not with round touchscreens—are the Tissot T-Touch and Silen-T wristwatches [28], shown in Figure 1. Both watches feature a “push-to-touch” mechanism, designed to avoid accidentally activating functionality. The T-Touch is activated by holding a finger to the touch-sensitive crown for longer than one second, and the Silen-T is activated by pressing in the crown (which is a button). This push-to-activate mechanism is a recurring feature of mobile systems in much of the literature. (a) (b) Figure 1: The Tissot T-Touch (a) and Silen-T (b) touch-sensitive watches. The Silen-T’s touch mechanism is activated by touching the crown (next to the E on the bezel) for longer than one second; the areas on the crystal labeled THERMO, METEO, ALTIMETER , CHRONO, COMPASS , and ALARM then become sensitive, activating various functionality when touched. The Silen-T also features a touchsensitive crystal, which is activated by pressing the crown. The watch then vibrates when the user’s finger is run clockwise around the face, with a long vibration when the finger is at the current hour position and short vibrations at the minute position. Although round touchscreen watches are absent in the literature, there has been work on rectangular touchscreen watches. IBM’s Linux-based WatchPad prototype (Figure 2(a)) incorporates a four-zone touchscreen, and Raghunath et al. implemented some simple interfaces for it [26]. However, because of the low resolution of the touchscreen, most of the applications depended upon the scrollwheel included on the watch (visible in Figure 2(a) on the left side of the watch). Blaskó further extended the IBM work, creating a rectangular prototype—although non-watchbased—with nine touch-sensitive zones (Figure 2(b)) that allowed for virtual scroll wheels [10], sliding, and gestures [9]. In his interfaces, Blaskó used the bezel at the edges of the display as a “tactile landmark” to inform the user in an eyes-free manner of their current location on the touchscreen. Unfortunately, the interfaces were only evaluated in stationary situations, and no consideration was given to the potential issue of accidental activation. One commercial (although not commercially successful) example of a touchscreen watch was the Fossil/Abacus WristPDA, which was based on PalmOS and had a 160x160 touchscreen. While its main menu showed only four icons (Figure 3(a)) and could conceivably be operated with a fingertip, other application screens were tiny enough (Figure 3(b)) to require the use of a stylus built into the watch band (Figure 3(c)). While not wristwatch-based, there have been a number of publications on the general topic of (a) (b) Figure 2: The IBM WatchPad (a) and Blaskó’s interface for a variation with nine touch-sensitive areas. circular interactions. Several techniques investigate menu selection; the most well-known are the Pie Menu [5], which uses a round menu with wedge-shaped buttons (Figure 4(a)), and its extension, Marking Menus [14]. These works have been extended in numerous ways: FlowMenu [11], for example, allows in-place hierarchical menu selection, command parameter entry and text input; earPod [38] used a pie-menu like design for an auditory menu (Figure 4(b)); and Schmidt et al. [29] developed a pie menu-like interface for wearable computers (Figure 4(c)). Scrolling is another common interaction using circular interfaces. The Radial Scroll Tool [30] allows a user to scroll through a document by making a “dial turning” motion. The Virtual Scroll Ring [21] is a similar idea. As well as moving through documents, scrolling widgets can be used for parameter selection: both FlowMenus and the Curve Dial [31]—the successor to the Radial Scroll Tool—allow for eyesfree parameter entry. Text entry has also been implemented using circular interfaces. FlowMenu uses an adaptation of the Quikwrite technique [25]. Cirrin [17] uses a circular key layout and a continuous gliding-style interaction to enter text (Figure 4(d)). Much of the literature on circular-style interfaces defines, whether explicitly or implicitly, two regions of interaction: the rim and the center. While these areas also make sense for a round touchscreen watch, I have encountered a lack of data on appropriate widget sizes for wristwatchsized screens. 2.3 Gesture Watch Activity recognition in general, and gesture recognition specifically, have been areas of research for a number of years. With the advent of inexpensive accelerometer chips, gesture recognition has moved from primarily computer vision-based to acceleration-based, with some systems [4] also incorporating MEMS-based gyroscopes. Gesture recognition may be differentiated from activity recognition by looking at the assumed intent of the input. Activity recognition is, in general, concerned with automatically determining what a user is doing at any given moment. Gesture recognition, on the other hand, assumes intentional input by the user with the explicit understanding that the movements made by the user will be used to perform some action. In this proposal, therefore, I will discuss only gesture recognition. Projects in gesture recognition can be divided into a spectrum based on the type of gestures (a) (b) (c) Figure 3: The Fossil/Abacus WristPDA. (a) shows the application menu, with icons large enough to push with a fingertip; (b) shows the calculator application with tiny buttons; (c) illustrates the scale of the included stylus that is normally tucked into the watch band. It is likely that the stylus or another pointed instrument would be needed to effectively use the calculator application. to be recognized. At one end of the spectrum lies sign language recognition. Starner et al. used computer vision to recognize a fixed set of American Sign Language (ASL) [33], and that work has been continued by McGuire et al., who incorporate accelerometers and recognize a much wider range of ASL [19]. The other end of the gesture recognition spectrum is occupied by fast, simple command gesture systems. Perhaps the simplest example is that of tapping or hitting a device in order to perform some action [27]. More complicated interactions can be achieved by moving towards the language end of the spectrum. One of the earliest of this type of interfaces was the Gesture Pendant (Figure 5(a)), a project I worked on in 1999 [1]. The Gesture Pendant was a computer-vision based system that allowed broad gestures to be used to control home appliances. The functions controlled by the gestures were largely parameter-adjustments, such as changing the volume on a stereo, the brightness of a light, or the channel on a television. The FreeDigiter [20] system was a derivative of the Gesture Pendant, using infrared proximity sensors to recognize splayed fingers passing in front of the mounting position on headphones (Figure 5(b)); this could be used to control a mobile phone, mobile audio player, or other device. Amongst the improvements made on the Gesture Pendant was that the proximity sensors allowed for a much smaller size and unobtrusive design. Proximity sensors were also used for another Gesture Pendant derivative, the Gesture Watch [12]. The Gesture Watch placed the proximity sensors on the wrist (Figure 5(c)), allowing the user to make gestures in the air above the device. One issue shared by the Gesture Pendant and its derivatives is how to tell when the user has intended to initiate a gesture. The approach taken by these projects has generally been to take a “push-to-gesture” approach. Similar to “push-to-talk” for speech recognition systems, push-togesture has the user push a button (Gesture Pendant) or make a known gesture (FreeDigiter, Gesture Watch) to initiate and end gesture recognition. In Chapter 5, I propose a solution to this commonlyencountered problem. (a) (b) (c) (d) Figure 4: Interfaces using a center/rim style of interaction. (a) is an example of a pie menu [5], (d) illustrates Cirrin [17], (b) is a conceptual visualization of the earPod menu system [38], and (c) is a pie menu-like selection system designed for see-through head-mounted displays (the central black area will appear to be transparent to the wearer). (a) (b) (c) Figure 5: Three on-body command-gesture devices. Device (a) is the Gesture Pendant [1], (b) is an example of FreeDigiter [20] in use, and (c) illustrates the Gesture Watch [12]. CHAPTER III MICROINTERACTIONS Microinteractions, as discussed briefly in Chapter 1, are interactions with a device that take a short time to complete. I gave a number of examples in that Chapter, that for convenience I will replicate here: checking the time, skipping to the next track on a digital audio player, sending a call to voicemail, reading a text message, checking the arrival time of the next bus, taking down a phone number, getting an idea of appointment times for the day, dialing a phone number from an address book, calculating a tip, checking the weather, updating status information in a social notification service such as “Twitter”, and setting or responding to a reminder alarm. These functions are clearly desirable—they serve to make everyday life easier and reduce the cognitive burden of many activities, and all are supported on many current devices in one form or another. Most of these tasks, however, are hidden behind layers of interface that the user must navigate through in order to reach the requisite application. In addition, devices are often carried in a pocket, purse, or bag, making access a cumbersome task at best. I define setup time as the time taken to get to the point in use of a device that the desired task can be accomplished. As will be shown, setup time can be divided into access time—the time taken to access and make ready the device itself—and navigation time—the time taken to navigate through the interface of the device until the desired function can be performed. To show that in microinteractions the setup time is important, I will present two pieces of evidence. The first is from a calendaring device study performed by several members of our lab in 2004 [35]. The second is a study I recently completed and that has been accepted for publication to CHI [2]. The first study (Section 3.1) indicates that if a device takes too long to access, people may simply not use it at all, preferring instead to rely on a faster method. The second study (Section 3.2) quantifies access time for three device placements in two different mobility conditions. Section 3.3 offers some discussion on how these two studies impact my proposed thesis work of reducing both access and navigation time. 3.1 Calendar Study This section uses data from a study conducted in early 2004 by Thad Starner, Neils Snoeck, Ben Wong, and Marty McGuire [35]. The study surveyed 138 participants (88% age 18-25, 70% male, 90% students) passing through Georgia Tech’s Student Center as to their preferred method of scheduling appointments when away from their desks. Participants were then asked to schedule an appointment with the researcher, and the method actually used was noted. Device Scrap Planner PDA Retr 17.8 11.8 11.0 Nav — 7.6 12.7 Retr + Nav 17.8 19.4 23.7 Entry 18.1 12.5 14.0 Total 35.9 31.9 37.7 Table 1: In seconds, average time for events during scheduling appointments, per commonly used device. Navigation time for scrap paper is effectively zero. 11 Memory Scrap Planner PDA (0s) (17.8s) (19.4s) (23.7s) Memory 24 1 — — Scrap 9 13 — — Planner 16 13 14 — PDA 4 1 1 8 Table 2: Claimed (columns) versus actual (rows) usage of devices for scheduling task. The rows are sorted by retrieval + navigation time, from Table 1, shown in parentheses. Four devices were commonly used to perform scheduling: the participant’s own memory, a scrap of paper, a planner, and a PDA. Note that at the time the study was conducted, smartphones were still uncommon. Table 1, adapted directly from the paper, shows the mean time per device for retrieving the device, navigating through the interface to the proper place to schedule the appointment, and the time to enter the appointment. Table 2, also adapted directly from the paper, sorts the four devices plus memory by retrieval plus navigation time (I have elided the “other” column and row from the table, as no timing information is given in the paper for this category). It shows the device that the participant claimed that they used for scheduling versus the device that was actually used when the researcher tried to schedule an appointment with the participant. The table shows that when a participant failed to use the claimed device, they almost always picked a method of performing scheduling that allowed them to get to the device faster. 3.2 Quickdraw This Section is adapted from [2], accepted to CHI 2008. It provides quantitative data about the amount of time required to access a device in three positions on the body, and in two mobility conditions. 3.2.1 Introduction Mobile phones have become ubiquitous, with the number of subscribers worldwide predicted to pass three billion in late 20071 . It is not uncommon for mobile phones to be the first object with which people interact in the morning and one of the last things with which people interact in the evening [7]. Despite the importance of these mobile devices in everyday life, little empirical data has been published on many fundamental usage properties. One of these underreported-upon properties is access time—the amount of time it takes a user to get to a device for the purpose of interacting with it. To take the mobile phone as a familiar example, the access time is how long it takes to get the phone from its storage place—such as a pocket, holster, or bag—and become ready to interact with the device by orienting it properly. In this work, we examine and quantify two important factors that could impact access time: onbody placement and user mobility. Specifically, we focus on the effects of walking versus standing when accessing a touch screen interface mounted on the wrist, on the hip, and in the pocket. This work is not intended to influence users to change their behavior with respect to mobility or device storage, but to allow designers a better understanding of how mobility and placement affect access time. 1 http://www.portioresearch.com/WWMM stats07.html 3.2.2 Experiment To explore how body placement and mobility influence the time needed to respond to and access a mobile device, we examined two independent variables with a 2x3 Latin-square within-subjects study design. The first variable is the mobility of the participant and the second is the placement of a device on the body. Although our interest in mobile devices is broader than telephones, we used a mobile phone throughout the study in order to keep the interaction consistent between conditions. Our two mobility conditions are standing and walking, chosen because people on-the-go are likely to be in one of these two states much of the time. Our three on-body placement conditions were in the pocket, on the hip in a holster, and on the wrist, reflecting common placements for current mobile electronics. During the standing condition we instructed participants to stand in a corner of our lab to minimize visual distractions that might interfere with access time. For the walking condition, participants were instructed to walk at a normal pace around a track constructed in our laboratory (Figure 6). The track was approximately 26 meters long and was denoted with flags hanging from the ceiling with the tips 0.75 meters apart. Each flag was hung so the tip was approximately 1.6 meters above the floor. We chose to use this flag arrangement rather than floor-based track markers (as in [36]) because walking is in general a head-up task (that is, people usually look ahead some distance while walking rather than looking directly at the floor). The walking direction was counterbalanced between clockwise and counter-clockwise directions. Figure 6: The path participants walked, starting at flag 1 and proceeding either clockwise or counterclockwise. For the device placement condition, we instructed participants to put the phone into a pants or skirt pocket (after removing other items), into the manufacturer-provided holster clipped to the top of the pants or to the belt, or to attach it to the wrist with a velcro strap (Figure 7). While using an actual touchscreen watch—such as SMS Technology’s M500 mobile phone watch—would have allowed for a more natural experience, we opted instead to maintain internal validity by using the same device (and therefore retaining the same display and touch input properties) for all three conditions. To determine the amount of time required to access the phone, we asked each participant to perform a simple task. Periodically, the phone generated an alert in the form of a sound and vibration. Each participant was requested to respond to these alerts as quickly as possible. When the alert occurred, the participant retrieved the device and looked at the screen, which showed a blue box and a large number (Figure 8(a)). The participant made a mental note of the number on the display and slid the blue box to the right to unlock the phone. The participant then chose the number they had just seen from a list of four numbers 8(b). After choosing the number, the participant returned the phone to its original position and waited for the next alert. (a) (b) (c) Figure 7: Placement: pocket (a), hip (b), and wrist (c). 3.2.2.1 Software and Equipment The software for our study was implemented in Python on a Motorola E680i cameraphone running the GNU/Linux operating system. The E680i has a 320x240 color touchscreen, stereo speakers, and a number of buttons. For this study, we used only the touchscreen and deactivated all of the hardware buttons so they would not be pushed accidentally. Pygame, a Python interface to the graphics library SDL, was used to create the user interface and to log user actions. During each condition, the operation of the software was the same. At random intervals between 20 and 40 seconds (selected from a uniform random distribution), the software generated alerts. An alert consisted of a loud mobile phone ringing sound and vibration that lasted up to 20 seconds. During the alert the software also displayed the prompt number and unlock mechanism on screen (Figure 8(a)). To respond to the alert, participants first unlocked the phone. Unlocking was accomplished by sliding the blue box to the right edge of the screen (similar to the Apple iPhone unlock gesture). This mechanism was implemented to prevent accidental responses while retrieving the phone from the pocket or holster. Next the phone displayed a screen consisting of a two by two grid of numbers (Figure 8(b)). The software waited for the user to select a number and logged the response. This interaction emulated common interruptions on mobile devices (for example: receiving a call, reading the caller ID, and sending the caller to voicemail). At this point the trial was complete. The phone relocked itself, and a timer was set to generate the next alert. The software logged the timestamps of each alert, the movements of the slider, and the selection of the numbers. Because the phone had no mechanism for determining whether it was in a pocket or holster, we implemented an extremely simple light sensor using the built-in camera. Approximately four times per second, the pixel values from the camera were summed and stored. With the assumption that the holster and participants’ pockets would be dark, we could detect when the phone was removed from, and replaced in, the pocket or holster. Figure 9 shows the output of the light sensor and other events logged by the software. (a) (b) Figure 8: Screen (a) is shown when an alert occurs. The participant must mentally note the displayed number and slide the blue box from the left to the target box on the right. Screen (b) is shown after the slide is completed; from the displayed four numbers, the participant must touch the number that was displayed on screen (a). 3.2.2.2 Dependent Measures We are most interested in how device placement and mobility influences access time—the amount of time it takes for a participant to retrieve the device and respond to an alert. Figure 9 shows a timeline of a typical notification-response cycle. The points in the timeline are as follows: 1. 2. 3. 4. 5. 6. 7. Blank screen; participant walking track or standing. Alarm starts ringing. Participant pulls phone from pocket or out of holster. Light level increases to nearly 100%. Participant starts to move slider. Screen with four numbers is displayed. Participant has picked a number; screen returns to blank. Participant returns phone to pocket or holster. Light level falls to 0%. We extracted several measurements from the timeline (numbers refer to the timeline in Figure 9): • Access time: Alarm start (2) to user acknowledgment of the alarm by moving the slider (4). • Pocket time: Alarm start (2) to retrieval of the device from the pocket or holster (3) (does not apply to wrist). • Hand time: End of phone removal from the pocket or holster (3) to the beginning of participant’s response (4) (does not apply to wrist). • Slide time: Moving the slider (4 to start of 5). • Answer time: Picking a number from the set of four (5). • Replacement time: Replacing the phone to the pocket or holster (start of 6 to 7—does not apply to wrist). Figure 9: Timeline of events and status of brightness detector during one notification-response cycle. The top line is the percentage of light detected from the camera (complete darkness, 0% on the bottom), and the bottom is the timeline of events recorded by the phone’s logging software. See the text for further details. We define access time in this context as the time required for the participant to react to the alarm, acquire the device (either looking at the wrist or pulling it from a pocket or the holster), note the number on the screen, and touch the slider to begin sliding it. 3.2.2.3 Procedure The evaluation for each participant began with the researcher presenting an overview of the study. Participants completed a consent form and a short survey about their use of mobile technology. The researcher then explained the experimental software and tasks. Each participant practiced responding to alerts three times on the phone for each of the placement conditions (pocket, hip, wrist) resulting in a total of nine practice responses. Next, the researcher familiarized the participant with the track by walking it twice in each direction; on the first lap the participant followed the researcher, while on the second, the researcher followed the participant. Finally, the researcher answered any questions from the participant, and started data collection. Each condition consisted of a set of trials. The number of trials per condition was either five or seven, averaging to six per condition per participant. This design was selected to prevent participants from anticipating the end of a set of trials. For the walking conditions, participants were told they could slow down or stop if needed to respond to the alert quickly, but to keep walking if possible. After the completion of the required number of trials, the phone displayed “STOP” in a large font on the screen. Participants were requested to stop where they were (if in the mobile condition) to allow the researcher to measure how far they had walked around the track. 3.2.2.4 Participants We recruited fifteen participants (5 females) for our study from our academic institution. The average participant age was 24.87 years (SD = 2.99). Fourteen of our participants were right handed. Each of our participants owned a mobile phone and all but three had that phone with them on arrival. Six of our participants wore a wristwatch when they arrived to participate in the study. We also asked about several other devices and where the participants carried them. Table 3 provides a summary. Mobile phone Phone headset Audio player Camera Total Pocket 7 4 4 6 21 Bag 4 — 5 5 14 Hip 1 — 1 — 2 Body — — 1 1 2 Total 12 4 11 12 39 Table 3: Devices carried by participants and their location. “Body” refers to on-body placement of a device, such as clipping to clothing or using a wrist strap. “Bag” includes purses and backpacks. 3.2.3 Results One participant was discarded as an outlier—having taken up to 8.5 standard deviations longer than average to respond to alerts—leaving 14 participants. Each participant performed all six of the conditions. Each condition averaged six alerts per condition, for a total of 504 alert-response cycles. Table 4 shows a summary of data for each condition. For this paper, we consider p < .05 to be significant and will only report p values for non-significant results. A multi-way ANOVA reveals that the placement of the device has a significant effect on access time, but mobility is not significant (p = .14). There is also no significant interaction between mobility and placement (p = .81). Post-hoc analysis of the access times using a paired Student’s T-test reveals a significant difference between all three combinations of variables: hip/pocket, hip/wrist and pocket/wrist. Therefore, we have a total ordering of access time for the placement condition: wrist (M = 2.8) < pocket (M = 4.6) < hip (M = 5.5); by comparing the mean access times, we can see that the pocket condition yields a 66% longer access time than the wrist while the hip requires 98% more access time than the wrist. For non-watch conditions, where the light sensor was used, pocket time was significantly affected by placement, but not mobility (p = .128), and there was a significant interaction between placement and mobility. There was no significant effect of placement for hand time (M = 1.145, SD = .628), but there was a significant effect for mobility as well as an interaction. Some, although not all, of the other measures described earlier were significant (refer to Figure 9). The answer time was found to be significantly affected by mobility, but not placement (walking: M = 1.227, SD = .304; standing: M = 1.479, SD = .583). Measures non-significantly impacted by placement or mobility were slide time (M = .071, SD = .72) and replacement time (M = 3.195, SD = 1.787). Time Hand Pocket Access Time Placement Hip Pocket Hip Pocket Hip Pocket Wrist Walking 1.047 (.656) 1.109 (.647) 4.355 (.906) 3.427 (.770) 5.377 (.947) 4.414 (.904) 2.728 (.291) Standing 1.339 (.728) 1.093 (.396) 4.312 (.770) 3.829 (.788) 5.660 (1.155) 4.817 (.875) 2.846 (.420) Average 1.199 (.707) 1.092 (.536) 4.333 (.836) 3.625 (.868) 5.518 (1.047) 4.616 (.897) 2.787 (.360) Table 4: Statistics for mean (SD) hand, pocket and access times in seconds. 3.2.4 Quickdraw Discussion The watch placement condition resulted in much faster access to the device. This finding is unsurprising, because participants did not have to remove the phone from the pocket or holster in order to use it. In Figure 9, access time (from 2 to 4 in the timeline) is divided into two segments for non-wrist conditions: pocket time (from 2 to 3) and hand time (from 3 to 4). The statistics in Table 4 reveal that the majority of access time is consumed by pocket time2 : on average, 78% of the time from the alarm until the participant started moving the slider was involved in getting the device out of the pocket or holster! The watch condition also resulted in much more consistent access time to the device; the standard deviation of access time for the pocket and hip conditions is 2.5 and 2.9 times more than the wrist condition, respectively. The long time required to retrieve the phone from its holder may help explain why the participants in the Cui et al. study reported missing phone calls due to how they carried their phones [8]. While we anticipated the superior performance of the watch condition, we did not expect the holster to perform worse than the pocket condition for access time. Reviewing user comments made during the study, however, it becomes clear that poor holster design may account for this result. Despite our use of the manufacturer’s holster, several participants complained during practice trials that the holster was difficult to use, and that difficulty may have persisted through the course of the study. An additional possibility is familiarity; while participants were presumably wellpracticed with putting items into and removing them from the pockets of their own pants, none of the participants reported using a holster to carry their phone (Table 3) and therefore may have been slower with the holster than would have otherwise been expected. Given these factors, our results for the hip condition should probably be viewed as a worst case for holster access. Finally, it was interesting to find that, in contrast to previous work [36], walking did not significantly impede user performance relative to standing still. In contrast, our data was trending to show that standing resulted in slower access. One possible explanation for this difference is our inability to separate reaction time and access time with our current experimental design. It is possible that the walking conditions kept the participants more engaged in the experiment relative to the standing conditions, and therefore the reaction time was slower while standing. 3.3 Discussion The results from the calendar study in Section 3.1 imply that the setup time involved in retrieving and navigating within the scheduling device, if long enough, can dissuade a person from using the device at all. This conclusion is borne out by other work as well [6, 15]. This suggests that decreasing the access and navigation times may improve interaction with the device. The calendar study directly influenced further work that resulted in a project called “DualPurpose Speech” (DPS) [16]. DPS greatly reduced in-application navigation time for scheduling appointments by utilizing the user’s regular conversational speech as input to a wearable appointment system. When, in a conversation, the user said something like “Okay, Monday at 3:30 sounds good,” the system would automatically navigate to the correct day and time in the appointment software. Techniques such as DPS to reduce in-application navigation time are one way to increase the 2 Note that in Table 4 pocket + hand 6= access. Two participants had light sensor problems, making hand and pocket time impossible to recover. Thus, these subjects were not included in this subsection of the table, making the quantities not exactly sum. usability of devices, but have the weakness of being domain-dependent. To use DPS with a different application would require training the system for a new set of vocabulary, and figuring out how to unobtrusively mesh commands to the system with everyday speech for that application. Reducing access time for a task is an alternate approach that may be easier to implement. The “Quickdraw” study described in Section 3.2 looked at how different on-body placements of a device might affect access time for the device. By placing a device on the wrist, access times were greatly reduced. Good placement is one component to reducing overall setup time, and DPS-like techniques reduce in-application navigation time. However, the remaining issue is that of out-of-application navigation time: getting to the application or function that is desired. This thesis proposal focuses on two methods for decreasing this time. CHAPTER IV TOUCHWATCH As discussed in the Introduction, touchscreens are a well-understood and common method for interacting with mobile devices. Several touchscreen watches have existed (see Chapter 2) but most have very simplistic input capability—such as the IBM WatchPad—or require the use of a stylus for more complex input, as with the Fossil WristPDA. Neither of these methods is appropriate for on-the-go microinteractions. If a user wants to quickly dial a phone number for a call on a Bluetooth headset, a stylus will take a frustratingly long time to access, and trying to input a phone number with only four soft buttons is likely to be time consuming as well. For these reasons, I want to explore how a finger-only touchscreen can be used for quick interactions with a touchscreen-based watch. As discussed in Chapter 2, Blaskó has performed some research in this area, especially with respect to manipulating numeric parameters. However, his touchscreen watch is rectangular—as have been all other touchscreen watches I have been able to find thus far. This rectangular form is somewhat at odds with standard wristwatch design. The convention of circular watch displays has a long history tracing to some of the first wristwatches, pocket watches, and the first analog clocks [18]. This design has become so ingrained in our culture that even many digital watches still have round faces surrounding the blocky digital time readout. Given this apparent cultural preference, I propose to investigate what touchscreen interactions should be like for a watch with a circular display. One approach to round touchscreen interaction is to follow the convention of other circular interfaces and put buttons around the edge. I conducted an experiment to determine what the best way for a user to interact with such a rim-based interface might be, and to find out with what density buttons could be placed around the edge. The results are presented in Section 4.1, which is partially excerpted from a paper in preparation for submission to MobileHCI 2008: D. Ashbrook, K. Lyons, and T. Starner, “My Watch is on the Fitts: an Investigation into Round Touchscreen Wristwatches”. Section 4.2 follows with some discussion on how to use the results to design an interaction technique, and Section 4.3 proposes a study to evaluate the effectiveness of this technique. 4.1 Experiment We conducted a Fitts’ Law experiment [32] to investigate the feasibility of interactions on a small, round touchscreen watch. We are particularly interested in the properties of an interaction area around the rim of the watch face. 4.1.1 Variables Movement type is our primary independent variable. Considering the kinds of movements that might be used for selection on the face of a touchscreen watch, we decided to test three methods. A tap to select is the simplest method, much like using a mouse to click on icons. This is the tap condition. Another possibility is sliding the finger from one place on the surface to another; this interaction is used in variety of interfaces such as Cirrin [17] and FlowMenu [11]. There are several possibilities for sliding on a watch surface. A natural method is to pick the shortest distance between the finger and the target, sliding the finger through the center region of the face; this forms 20 the through condition. Another option is sliding the finger using the bezel as a guide, avoiding the center region of the watch. This is the rim condition. The main dependent variable of interest is the effective target width, We [32]. This measure indicates the size of a target that would accommodate 96% of the attempts to select it under Fitts’ Law conditions. Calculating this number will give insight into how many target regions, i.e. buttons, could be placed around the rim of the watch, and how much space will be left for a display in the center. The calculation of We is discussed in Section 4.1.6. 4.1.2 Design Our experiment is a repeated measures within-subjects design. The primary independent variable of interest is movement type (tap, through, rim). We also experimentally manipulated the distance between targets to obtain data needed for our Fitts’ Law calculations. Because our surface of interaction is circular, we used degrees to specify the target placements and distances. In order to ensure that our data covers the face of the watch, we first randomly chose a start position from a uniform random distribution from 0◦ –360◦ . Using a time metaphor, we next divided the face of the display into twelve segments of 30◦ each. We further divided each segment in half to get a minimum distance between targets of 15◦ . We used all of the distances in 15◦ increments between the minimum of 15◦ and a maximum distance of 180◦ , resulting in a total of twelve different distances. Note that for the tap and through conditions, movement are straight-line, so a 180◦ distance will yield a movement equal to the diameter of the watch—1.25in (30mm). The rim condition, however, will yield an arc length, so a 180◦ distance will actually yield a movement equal to πr, or 1.96in (47mm). (a) tap (b) through (c) rim (d) tap (e) through (f) rim (g) key Figure 10: Images (a), (b) and (c) show the targets (short green lines at rim of circle) and movement guides (yellow line between targets) displayed to study participants. Images (d), (e) and (f) show actual movements for one session; blue lines indicate movement by participant’s finger and small yellow circles show inflection points. Movement images were displayed only to researcher. Images (a) and (d) are for tap condition; images (b) and (e) are for through condition; images (c) and (f) are for rim condition. Images (a)–(f) are printed at actual size. Figure 11: The simulated touchscreen watch. The experiment was composed of numerous trials. For each trial, the participant moved back and forth between the pair of targets 15 times, resulting in 30 movement end points. A block is a set of twelve trials (one per distance) with one movement type (e.g. rim). The blocks were repeated three times for each of the three movement types. Therefore we have 12 distances × 3 repetitions × 3 movement types = 108 trials per participant. The movement types and distances were ordered according to a Latin square to minimize potential ordering effects. In between each block, participants were given an enforced 30-second break to avoid fatigue; if desired, they could break for longer than 30 seconds. In order to avoid biasing participants with any particular size of target, and lacking recommendations from the literature, we selected targets as close to zero width as we could display. These were simple lines extending from the outside of the watch towards the center (Figure 10). Each line was one pixel wide, or approximately 0.58◦ at the inner end of the target. While these targets are unrealistically small, they will allow us to determine the effective width We of a target by considering the distribution of nearby hits. Using the Shannon formulation for the index of difficulty, D +1 (1) ID = log2 W this small target size yields a maximum range of ID values from 4.64 to 8.3 bits. 4.1.3 Apparatus Lacking a programmable round touchscreen watch, we simulated one. We removed the case from a Motorola E680i touchscreen mobile phone and placed it into a custom-made plastic case (Figure 11). In order to simulate a round watch face with a bezel, we placed over the screen a customcut plate of .06in (1.6mm) thick steel with a hole approximately 1.25in (30mm) in diameter, with a sloping bevel around the hole. The “watch” was mounted on each participant’s arm using a velcro strap, with the visible portion of the screen centered on the arm approximately where a normal wristwatch face would be. Software was written for the phone using Python and Pygame (an interface to the SDL graphics library) to present trials to participants and collect data. 4.1.4 Procedure Participants performed a Fitts-style reciprocal 2D pointing task using their finger as the pointer. Two targets were displayed on the watch face and participants were requested to select the targets as quickly and accurately as possible. For sliding conditions, a line was drawn to remind the participant what kind of motion to make (Figure 10). Because we were trying to determine We in addition to standard Fitts metrics, we did not force participants to hit the targets—instead, our software simply detected the endpoints of the movements. In the tap condition, the participant lifted their finger in between each tap of a target. In the through condition, the participants were instructed to slide their finger along the shortest distance between the targets. The rim condition was designed with bezel-based interactions in mind, and required participants to slide their finger along the rim of the watch between the targets, making an arc. 4.1.5 Participants Fifteen volunteers were recruited to participate in the experiment. Fourteen volunteers were male and one was female; ages ranged from 20 to 28 (SD = 1.8). Participants were paid US $10 per hour for participation; no participant took more than one hour to complete the experiment. All participants were either right-handed or primarily used their right hand for control tasks such as mouse movement. Eight participants reported normally wearing a watch, and all stated they normally wore the watch on the left wrist. 4.1.6 Results The standard deviation σ of each cluster was calculated, and points falling greater than 3σ away from the cluster mean were discarded. We calculated the effective target width We [32]. This metric is defined using the standard deviation of the endpoints of movements: We = 4.133σ. Statistically, for a button of width We , 96% of attempts to hit the button will be successful. We calculated the effective width both parallel and—while not a traditional measure—perpendicular to the direction of movement. For tap and through, this meant along and perpendicular to the line between the cluster means; for rim, this was around the circumference of the circle and on the line from the center of the circle to the center of each mean. The mean value of We (in pixels) for each condition is listed in Table 5. Condition tap through rim We k 25.1 (5.0) 22.9 (5.2) 22.9 (7.8) We ⊥ 25.3 (5.5) 22.4 (5.9) 10.1 (3.8) Table 5: Mean (SD) effective width We in pixels for each movement condition, parallel (k) and perpendicular (⊥) to the direction of movement. The effective distance De was calculated as the mean of the distances between each pair of movement end points, which in turn allowed the effective index of difficulty IDe to be calculated. This is similar to the index of difficulty ID, except that the effective distance De and effective width We are used instead (see Equation 1). The range of IDe values for each condition were: 0.90–3.66 (M = 2.32) bits for tap; 0.63–4.44 (M = 2.40) bits for through; and .74–3.97 (M = 2.26) bits for rim. Figure 12 shows a plot of the effective index of difficulty IDe versus the movement time MT , and shows corresponding regressions for each movement condition. Finally, the throughput values were calculated. Throughput combines both speed and accuracy, and is calculated as T P = IDe /MT . For tap, the mean T P was found to be 7.7 (SD = .98) bits/sec; for through, 11.8 (SD = 1.0) bits/sec; and for rim, 7.2 (SD = .72) bits/sec. (a) tap (b) rim (c) through Figure 12: Effective index of difficulty (IDe ) versus movement time (MT ) for the three movement conditions. The regression line is dotted in each graph. Regression statistics are: tap slope: 40ms/bit, intercept: 228ms, R2 : 0.04); through slope: 54ms/bit, intercept: 106ms, R2 : 0.09); rim slope: 158ms/bit, intercept: −18ms, R2 : 0.45). 4.1.7 Discussion Our tap and through conditions are typical Fitts’ tasks, though rim is not. However, Figure 12(c) shows that the correlation between effective index of difficulty IDe and movement time MT is reasonably linear (R2 = .45) for rim. The R2 values for tap and through—.04 and .09, respectively— do not show as close a fit; we believe this is due to the small range of IDe values. Fitts’ original paper found a range of IDe values between 1 and 6.02 bits [32], while we have the smaller range of 0.63—4.44 bits. Our effective indices of difficulty can be attributed to several factors. The first is the small size of the watch face—only 1.25in (30mm) in diameter, giving only a relatively small range of possible movement distances. The second factor is our small target size. By giving only thin lines for targets and not enforcing via software the participants hitting the targets, we essentially allowed participants to choose their own widths and distances, growing We and shrinking De , therefore decreasing the range of IDe values. Another possible influence is that for some of the smaller distances, a participant’s finger could occlude the targets, removing the visual guidance necessary for a Fitts’ task. Given the effective width values for the three conditions, we can now think about what a round touchscreen watch interface might look like. Our initial approach was to divide the circumference of the watch face by the effective width for a particular movement type and put that many buttons around the edge. However, we discovered that participants, possibly due to the width of their fingers, did not touch the touchscreen close to the edge: the mean distance Dµ from the rim for tap was 24.3 (SD = 10.8) pixels; for through 21.8 (SD = 10.8) pixels; and for rim 18.3 (SD = 8.7) pixels. To determine the optimal number of buttons to place around the rim, we can use Dµ , the mean distance from the rim. Dividing the circumference of this circle by the effective width gives the maximum number of buttons: 2π(r − Dµ ) N= We (a) tap (b) through (c) rim (d) key Figure 13: Possible watch face layouts. Where r is the watch radius and the We used is parallel to the movement direction for rim and perpendicular for tap and through. As illustrated in Figure 13, this calculation yields 15 buttons for tap, 18 buttons for through, and 20 buttons for rim. In Figure 13, images (a)–(c) are printed at the actual size of the physical watch face. In the key, for tap and through, We1 represents We perpendicular to the direction of movement and We2 is parallel; for rim these are switched. The number in the center of each layout is the percentage of screen area that is left for the center region after the buttons have been placed. 4.2 Rim Interfaces How may the results of the experiment be used to design interfaces? Although the rim method resulted in the best performance, it is bound to be unfamiliar to novice users. If simply presented with 20 buttons around the watch rim, it is likely that novices would use something similar to the tap method to interact, causing errors and frustration. Therefore, some method must be used to encourage users to utilize the most effective interaction. A second potential problem that may be encountered when using a touchscreen watch is erroneous activation. An accidental brush of the watch against an object or even a finger touch while adjusting the watch on the wrist could activate unwanted functionality. Both of these problems may be addressed by considering the round face of the watch. A round interface naturally lends itself to circular movements. By creating the interface such that users naturally use circular, rim-based movements, accidental activations will be reduced. However, how can such interfaces be created without feeling unnaturally forced to the user? (a) (b) (c) Figure 14: Blaskó’s card dragging technique [9]. The default watch face is illustrated in (a), with the asterisk denoting the desired card. The motion made by the user’s finger is illustrated in (b), and the resulting screen with the activated content is shown in (c). 4.2.1 Card Dragging In his thesis [9], Blaskó proposed the idea of pullable “content cards”, illustrated in Figure 14. Each card contains information or an applications, and “protrudes” slightly into the face of the watch from the edge. When the user wants to access a particular card, they drag the card from the edge down, “pulling” application or information into view. The benefit of this method is that it trained users in Blaskó’s on-screen gesture system. I propose to use a similar method to encourage users to use the rim method of interaction. Each button will represent a “card” containing information or an application. Users will, starting with the button, drag around the rim 360◦ (or so) to bring up the application. This interaction has a similar function to the iPhone’s slide-to-activate mechanism: it is exceedingly unlikely that a full 360◦ movement will be made accidentally, so any such movements can be treated as intentional on the part of the user. Figure 15 illustrates a mock-up of this idea, showing a theoretical interaction in which a user selects the weather application from a list and drags its “card” out of the “stack” of applications. (a) (b) (c) (d) (e) (f) Figure 15: Proposed “card dragging” interaction using icons on rim of watch. (a): view of watch with no applications active; (b): user touches face to show application icons; (c): user selects desired application icon; (d)–(f): user drags application “card” out of “stack”. Images are shown at actual size for hardware used during Fitts’ study and conform to design recommendations discovered therein. 4.2.2 Calendar Dialing Opening applications using rim dragging avoids false detections, but using other interaction techniques within the applications violates the consistency principle of HCI. In addition, having up to 20 possible buttons might be useful within applications. Therefore, I am interested in how the rim paradigm of interaction may be used in other applications. In the real world, one common interface that uses circular movements is the knob. This suggests that a knob metaphor may make sense on a round touchscreen watch as well. An already-familiar use of this metaphor is the Apple iPod’s touchwheel interface: the touchwheel is used for scrolling in lists and for changing the volume of playing audio. One commonality among most knobs is that of directionality—turning the knob clockwise increases a value or moves in one direction while turning it counterclockwise generally has the opposite effect. This idea is illustrated in Figure 16 for a calendaring application. Clockwise and counterclockwise turns are often mentally associated with increasing and decreasing; for example, clockwise turns of dials increase volume on stereos while counterclockwise turns decrease volume. The mockup uses this same metaphor of increasing with clockwise dialing and decreasing with counterclockwise dialing, but to zoom in and out of a calendar. (a) (b) (c) (d) Figure 16: Mockup of a possible dial-metaphor interface for calendar navigation. Dark gray segments in (a)–(c) show “now”, thick red borders indicate area selected, and bands around rim in (d) indicate appointments. User touches red-banded area and drags finger clockwise to zoom in to next closer detail view; to zoom out, user drags finger counterclockwise. The lowest detail view, of the months of the year, is illustrated in Figure 16(a). The current month, June, is indicated by coloring that segment in dark gray. To zoom into a view of June, the user puts a finger on the “Jun” segment (this touch is indicated by a red border in the Figure) and quickly moves it 360◦ around the rim in a clockwise direction. The calendar zooms into a view of the weeks of June, shown in Figure 16(b). Although 28 days are shown (possibly with coloring or other indicators of schedules for that day), we learned in Section 4.1 that 20 is the maximum number of buttons that should be placed around the edge. Therefore, as an intermediate step, the user selects the entire week. This is accomplished in the same way as zooming into the month: the user quickly rotates a finger 360◦ clockwise around the rim. The display changes to a week view in Figure 16(c), and the same interaction can take place. Finally, 16(d) shows the day view with bands indicating schedule items. A user could further zoom into the calendar at this point—perhaps by selecting an appointment band—to edit information or view more detail. One can imagine extra information shown at each of the zoom levels to indicate number of appointments, upcoming schedule items, priorities, or other data. 4.3 Proposed Work The experiment described in Section 4.1 gives recommendations for the numbers of buttons for a round touchscreen watch interface, but the results have not been empirically validated. Therefore, I intend to perform a second study to validate the results. The procedure will be similar to the first study, but will be in the traditional Fitts vein, with targets of specific sizes and locations. Once the second study has been completed and the results from the first have either been validated or modified appropriately, I will create several prototype interfaces using the rim method of interaction to determine how effective it is. I will implement the application selection and calendar zooming interfaces described in Section 4.2 and compare performance and user preference to a standard-style interface such as on the Palm-based Fossil WristPDA. CHAPTER V WRIST-BASED GESTURE INTERACTION I propose to create and study a wristwatch prototype that performs basic gesture recognition. Rather than a complex, many-movement language such as American Sign Language, my proposed system will recognize a small collection of quick, unobtrusive gestures used for control rather than communication. I call my overall system MAGIC: Mobile Action Gesture Interface Control. 5.1 Gesture Triggering One issue with on-body gesture recognition is how to trigger the start of the recognition process. A common solution is to have a “push to gesture” mechanism, either a physical button [1] or other intentional action [3] that signals to the computer that a gesture is about to be performed. This increases the accuracy of recognition because there is much less chance of random movements being erroneously recognized as intentional gestures. This is especially important in a system intended for everyday use. The drawback to the trigger approach is that an extra motion must be made. In the case of a wristwatch—such as the Tissot (Figure 1)—it is not physically possible to push buttons on the watch with the hand belonging to the wrist that the watch is attached to. This necessitates the use of the other hand to push a button, which obviates the purpose of having control gestures rather than buttons in the first place. An alternative to using a button could be to use a starting gesture, after which the intended gesture is performed. This solution is superficially similar to wordspotting in speech recognition [37] (“Computer, save file”); however, such speech recognition systems are usually constrained to quiet office environments. In the context of everyday use, it becomes much more difficult to find a start gesture or word that is unobtrusive, easy-to-perform, and will have a low likelihood of mis-recognition. There is a solution to the triggering issue, however. By having each user create a database of movements during a typical day—an “Everyday Gesture Library” (EGL)—new gestures can be rated based on how likely they are to be mis-recognized. A user could purchase a device, wear it for two or three days, and then train it on a set of new gestures. For each candidate gesture, the system would then run through the EGL to see if it recognizes any of that data as the gesture under consideration. If it does, it would warn the user to choose a different gesture. Figure 17 illustrates the idea. A portion of an EGL recorded for one user is displayed in the top portion of (b), showing the accelerometer trace of some normal movements made during the course of a day. A candidate gesture—perhaps for a “play/pause” gesture for a digital audio player—is illustrated in (a). This candidate gesture is similar to a movement made by the user during a normal day; the similarity is indicated by the match percentage line shown in the bottom half of (b). The match percentage indicates there is a high likelihood of accidental triggering, so a different gesture should be chosen. This technique could be used in end-user devices, where the user would wear the device for several days to generate an EGL before training the system with a chosen gesture set, or could be used by interface designers using a larger EGL collected from a variety of representative users. A wide variety of interfaces that now use push-to-activate could benefit, from hand gesture-based 29 (a) (b) Figure 17: Candidate gesture performed by user (a) and partial gesture database (b) with match percentage (bottom) and database contents (top) illustrated. devices such as the Gesture Pendant [1] to touch-sensing interfaces like the Tissot T-Touch [28]. 5.2 Proposed work I propose to validate the idea of using an EGL to help pick the best gestures for an interface. While the EGL concept should apply to any gesture system intended for everyday use, in keeping with the theme of this proposal I intend to implement and test the concept with a wristwatch-based gesture system. 5.2.1 Hardware I am currently in the final stages of creating a wristwatch prototype with which to test this interaction concept. The watch will consist of a 1.58”/4cm (diagonal measure of visible area) organic LED (OLED) screen, with an ARM9 processor, 3-axis accelerometer, microSD card slot, and Bluetooth radio. By itself, the watch should be able to collect and store a day’s worth of movement data, but probably will not have the computational power to perform realtime gesture recognition. Instead, it will communicate over Bluetooth to a more powerful device, such as a mobile phone, that is nearby—perhaps in the user’s pocket. Figure 18 shows the two sides of the prototype hardware board. The board is 1.4”/3.56cm on a side. The screen is not shown in the photo, but sits on the processor side of the board with the cable wrapping around to the other side. 5.2.2 Software To perform the gesture database tasks discussed above, I have implemented proof-of-concept software. Server code runs on a Motorola e680i GNU/Linux-based mobile phone and records data from a wrist-mounted Bluetooth accelerometer. Separate software, written in Python and pyGame (a Python wrapper for the SDL graphics library), uses Dynamic Time Warping (DTW) [22] to provide visualization like that illustrated in Figure 17 and to perform both database searches and gesture recognition. Currently, the Python search and recognition code is somewhat slow: with graphics turned on (for feedback), it is slower than realtime; with graphics off, it is 4–5 times faster than realtime, but (a) (b) Figure 18: Prototype watch hardware. Features on the side of the board illustrated in (a), from the top left and in clockwise direction are the Bluetooth module, the power/charge switch, battery wires, reset and action buttons, display connector, and mini-USB connector (for charging and programming). (b) shows (in the same order), the microSD card slot, unpopulated pads for the accelerometer chip, and the processor. Also visible are programmable LEDs and Bluetooth status LEDs. this is still very slow when searching a 40 hour long database. I am working with JungSoo Kim, a MS student in our group, to convert the codebase to C, and to incorporate additional measures to improve performance, such as coarse-to-fine search using Gaussian pyramids. 5.2.3 Study In order to validate that creating and searching through an EGL is an effective way to determine a good set of gestures to use for a mobile interface, I want to conduct a study with interface designers. Several users will be given wristwatch hardware to record an aggregate data set over a period of several days. The data will be subdivided into two segments, one to form an EGL and one for testing purposes. A number of HCI students will be recruited to design gesture-based interfaces using the system. A variety of possible mobile interface functions—such as play/pause, volume up/down, or send to voicemail—will be selected, and participants will be asked to construct a series of gestures to activate these functions. The designers will be divided into a control group and an experimental group; the control group will design the gesture interfaces without the EGL system, while the experimental group will use the system. The gesture interfaces from each group will be evaluated for performance using the non-EGL segment of the collected data. The data will be used to simulate a user wearing the device in normal life, without trying to activate any gestures. The number of times that each of the designer’s gestures are erroneously recognized will be calculated to determine if using the EGL had a positive effect on the performance of the simulated device. CHAPTER VI PLAN FOR COMPLETION E XPECTED C OMPLETION TASK MAGIC (Chapter 5) Gesture hardware development Gesture software development Everyday Gesture Library study Writeup of paper for publication TouchWatch (Chapter 4) Mobile phone-based hardware mockup Button study software development Button size validation study Card dragging software development “Dialing” software development Interface study Writeup of paper for publication Thesis Thesis draft Defense 32 January 2008 January 2008 February 2008 March 2008 March 2008 March 2008 April 2008 May 2008 June 2008 July 2008 August 2008 September 2008 November 2008 REFERENCES [1] A SHBROOK , D., AUXIER , J., G ANDY, M., and S TARNER , T., “Experiments in interaction between wearable and environmental infrastructure using the gesture pendant,” in Proc. of HCII Workshop on Wearable Computing, (New Orleans, LA), 2001. [2] A SHBROOK , D., C LAWSON , J., LYONS , K., PATEL , N., and S TARNER , T., “Quickdraw: The impact of mobility and on-body placement on device access time,” in Human Factors in Computing Systems (CHI) 2008, 2008. [3] A SHBROOK , D., LYONS , K., and S TARNER , T., “My watch is on the fitts: an investigation into round touchscreen wristwatches,” in Submitted to Human Factors in Computing Systems (CHI) 2008, 2008. [4] B.V., X. T., “Xsens inertial motion sensor.” http://xsens.com, December 2007. http: //xsens.com. [5] C ALLAHAN , J., H OPKINS , D., W EISER , M., and S HNEIDERMAN , B., “An empirical comparison of pie vs. linear menus,” in Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI), 1988. [6] C AMPBELL , C. and M AGLIO , P., “Supporting notable information in office work,” in CHI ’03: CHI ’03 extended abstracts on Human factors in computing systems, (New York, NY, USA), pp. 902–903, ACM, 2003. [7] C HIPCHASE , J., P ERSSON , P., P IIPPO , P., A ARRAS , M., and YAMAMOTO , T., “Mobile essentials: field study and concepting,” in Designing for User eXperience, 2005. [8] C UI , Y., C HIPCHASE , J., and I CHIKAWA , F., “A cross culture study on phone carrying and physical personalization,” in HCI International, 2007. [9] G ÁBOR B LASK Ó, Cursorless Interaction Techniques for Wearable and Mobile Computing. PhD thesis, Columbia University, 2007. [10] G ÁBOR B LASK Ó and F EINER , S., “Evaluation of an eyes-free cursorless numeric entry system for wearable computers,” in Proceedings of 10th IEEE International Symposium on Wearable Computers (ISWC 2006), (Montreux, Switzerland), pp. 21–28, 2006. [11] G UIMBRETI ÉRE , F. and W INOGRAD , T., “Flowmenu: combining command, text, and data entry,” in Proceedings of ACM Symposium on User Interface Software and Technology (UIST), 2000. [12] K IM , J., H E , J., LYONS , K., and S TARNER , T., “The gesture watch: A wireless contactfree gesture based wrist interface,” in Proceedings of 11th IEEE International Symposium on Wearable Computers (ISWC 2007), 2007. 33 [13] K RISTOFFERSEN , S. and L JUNGBERG , F., ““making place” to make it work: empirical explorations of hci for mobile cscw,” in GROUP ’99: Proceedings of the international ACM SIGGROUP conference on Supporting group work, (New York, NY, USA), pp. 276–285, ACM, 1999. [14] K URTENBACH , G., The Design and Evaluation of Marking Menus. PhD thesis, Univ. Toronto, 1993. [15] L IN , M., L UTTERS , W. G., and K IM , T. S., “Understanding the micronote lifecycle: improving mobile support for informal note taking,” in CHI ’04: Proceedings of the SIGCHI conference on Human factors in computing systems, (New York, NY, USA), pp. 687–694, ACM Press, 2004. [16] LYONS , K., S KEES , C., S TARNER , T., S NOECK , C. M., W ONG , B., and A SHBROOK , D., “Augmenting conversations using dual-purpose speech,” in Proc. of ACM ACM Symposium on User Interface Software and Technology (UIST) 2004, (Santa Fe, NM), 2004. [17] M ANKOFF , J. and A BOWD , G. D., “Cirrin: a word-level unistroke keyboard for pen input,” in Proceedings of ACM Symposium on User Interface Software and Technology (UIST), 1998. [18] M ARTIN , T., “Time and time again: parallels in the development of the watch and the wearable computer,” in Proceedings of ISWC, 2002. [19] M C G UIRE , R., H ERNANDEZ -R EBOLLAR , J., S TARNER , T., H ENDERSON , V., B RASHEAR , H., and ROSS , D., “Towards a one-way american sign language translator,” in Proceedings. Sixth IEEE International Conference on Automatic Face and Gesture Recognition, (Washington, DC, USA), pp. 620–625, IEEE Computer Society, May 2004. [20] M ETZGER , C., A NDERSON , M., and S TARNER , T., “Freedigiter: A contact-free device for gesture control,” in Proceedings of 8th IEEE International Symposium on Wearable Computers (ISWC 2004), 2004. [21] M OSCOVICH , T. and H UGHES , J., “Navigating documents with the virtual scroll ring,” in Proceedings of ACM Symposium on User Interface Software and Technology (UIST), 2004. [22] M YERS , C. S. and R ABINER , L. R., “A comparative study of several dynamic time-warping algorithms for connected word recognition,” The Bell System Technical Journal, vol. 60, pp. 1389–1409, September 1981. [23] O ULASVIRTA , A., TAMMINEN , S., ROTO , V., and K UORELAHTI , J., “Interaction in 4-second bursts: the fragmented nature of attentional resources in mobile hci,” in CHI ’05: Proceedings of the SIGCHI conference on Human factors in computing systems, (New York, NY, USA), pp. 919–928, ACM, 2005. [24] PATEL , S., K IENTZ , J., H AYES , G., B HAT, S., and A BOWD , G., “Farther than you may think: An empirical investigation of the proximity of users to their mobile phones.,” in Ubicomp, pp. 123–140, 2006. [25] P ERLIN , K., “Quikwriting: continuous stylus-based text entry,” in Proceedings of ACM Symposium on User Interface Software and Technology (UIST), 1998. [26] R AGHUNATH , M. T. and NARAYANASWAMI , C., “User interfaces for applications on a wrist watch,” Personal and Ubiquitous Computing, vol. 6, pp. 17–30, February 2002. 34 [27] RONKAINEN , S., H ÄKKIL Ä , J., K ALEVA , S., C OLLEY, A., and L INJAMA , J., “Tap input as an embedded interaction method for mobile devices,” in TEI ’07: Proceedings of the 1st international conference on Tangible and embedded interaction, (New York, NY, USA), pp. 263– 270, ACM, 2007. [28] S.A., T., “Tissot wristwatches.” http://tissot.ch, December 2007. http://tissot.ch. [29] S CHMIDT, A., G ELLERSEN , H.-W., B EIGL , M., and T HATE , O., “Developing user interfaces for wearable computers: Don’t stop to point and click,” in Intelligent Interactive Assistance & Mobile Multimedia Computing, 2000. [30] S MITH , G. M. and M . C . SCHRAEFEL, “The radial scroll tool: scrolling support for stylusor touch-based document navigation,” in Proceedings of ACM Symposium on User Interface Software and Technology (UIST), 2004. [31] S MITH , G., M . C . SCHRAEFEL, and BAUDISCH , P., “Curve dial: eyes-free parameter entry for guis,” in Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI), 2005. [32] S OUKOREFF , R. and M AC K ENZIE , I., “Towards a standard for pointing device evaluation, perspectives on 27 years of fitts’ law research in hci,” Intl. Journal of Human-Computer Studies, vol. 61, no. 6, 2004. [33] S TARNER , T., “Visual recognition of american sign language using hidden markov models,” Master’s thesis, MIT Media Laboratory, Cambridge, MA, February 1995. [34] S TARNER , T. E., S NOECK , C. M., W ONG , B. A., , and M C G UIRE , R. M., “Use of mobile appointment scheduling devices,” in Proceedings of the SIGCHI conference on Human factors in computing systems, ACM Press, 2004. [35] S TARNER , T. E., S NOECK , C. M., W ONG , B. A., and M C G UIRE , R. M., “Use of mobile appointment scheduling devices,” in CHI ’04: CHI ’04 extended abstracts on Human factors in computing systems, (New York, NY, USA), pp. 1501–1504, ACM, 2004. [36] VADAS , K., PATEL , N., LYONS , K., S TARNER , T., and JACKO , J., “Reading on-the-go: a comparison of audio and hand-held displays,” in MobileHCI, 2006. [37] W ILCOX , L., S MITH , I., and B USH , M., “Wordspotting for voice editing and audio indexing,” in CHI ’92: Proceedings of the SIGCHI conference on Human factors in computing systems, (New York, NY, USA), pp. 655–656, ACM, 1992. [38] Z HAO , S., D RAGICEVIC , P., C HIGNELL , M., BALAKRISHNAN , R., and BAUDISCH , P., “Earpod: eyes-free menu selection using touch input and reactive audio feedback,” in Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI), 2007.