Paper
Transcription
Paper
A TETHERED AEROBOT FOR PLANETARY ROVER MISSIONS Laurence Tyler(1), Dave Barnes(1), Mark Neal(1), Fred Labrosse(1), Stephen Pugh(1), Gerhard Paar(2) & the PRoViScout Team (1) Department of Computer Science, Aberystwyth University, SY23 3DB, Wales, UK (2) Joanneum Research, Graz, Austria, A-8010 ABSTRACT We present an overview of the PRoViScout rover and tethered aerobot. The PRoViScout project involves visually-guided autonomous science target assessment, operations planning and navigation for a planetary rover class mission. A wheeled rover will carry a variety of primarily visual sensors with which to explore a Mars analogue site. Decisions about waypoints and science target selection will be made autonomously during the mission. A tethered aerobot provides area context for the rover in the form of multispectral imagery that is used to generate both a terrain model for navigation and a mineralogical map to inform science assessment and mission planning. Experiences and some results from early tests of the aerobot are presented. 1. The demonstrator system will comprise an autonomous rover platform carrying a number of vision-based sensors. Given an initial science plan and a Digital Elevation Model (DEM) of the area to be explored, the rover will conduct a fully autonomous scouting mission, including re-planning to allow for the opportunistic investigation of serendipitous science targets. The PRoViScout project is described in greater detail in [1], and see also [7]. INTRODUCTION PRoViScout is a project within the European FP7SPACE framework. It aims to demonstrate the utility and feasibility of equipping long duration planetary rover missions with vision-based systems for autonomous planning, terrain analysis and science target selection. The project will culminate in a field trial demonstration conducted in a Mars-analogue environment. Figure 2. Autonomous tethered aerobot Complementing the rover platform will be an autonomous tethered aerobot with a high-resolution multispectral camera. The aerobot provides vital area context for the rover, at a scale intermediate between ground observations and orbital imagery. Images captured by the aerobot will be used to generate an area DEM for mission planning and initial waypoint selection. The addition of mineralogical data derived from the multispectral bands of the aerobot images will allow further refinement of the initial science target selection. 2. Figure 1. Rover platform "Idris" AEROBOT RESEARCH AT ABERYSTWYTH The Space and Planetary Robotics group at Aberystwyth University have an ongoing interest and involvement in aerobot research. The group has taken part in several aerobot-based projects in conjunction with other partners. A previous project investigated autonomous imagebased localisation for a future Mars aerobot mission [2]. This was an ESA-funded project. Project partners included SciSys Ltd, University of Leicester and Joanneum Research. The project culminated in a demonstration of the developed localisation algorithms controlling a real aerobot. Fig. 3a shows the completed aerobot undergoing acceptance trials at the ESA ESTEC Planetary Testbed Facility. points downwards. The camera is mounted on a stabilising platform under the balloon along with its controlling computer, wireless network link and sensors for measuring the position and attitude of the aerobot. The projected ground view of the aerobot camera at an altitude of 100m is approximately 106m by 88m, with a resolution of 43 mm/pixel. This combines a useful level of local detail with broader area coverage for mission planning. The level of detail discernible is significantly higher than would normally be achieved by orbital imaging, and the wider surface coverage and elevated viewpoint of the aerobot is advantageous in traverse planning when compared to the more restricted view from the rover's own sensors. 4.1. Aerobot Hardware The prototype aerobot measures approximately 20cm x 18cm x 18cm and has a framework constructed from thin aluminium plate with nylon bracing bars. The aerobot is shown in fig. 4. 3a: Aerobot acceptance trials at ESTEC 3b: Cooperating aerobots flying in formation Figure 3. Aerobot research at AU AU has also investigated the use of cooperative control methods for the coordination of multiple aerobots [3]. This project demonstrated the developed control algorithms by autonomously flying a group of aerobots in formation in a large enclosed space (fig. 3b). Each aerobot operated its own behaviour-based controller. The cooperative control algorithm was able to maintain the formation pattern of the aerobots in the face of external perturbations. Project partner was SciSys Ltd. 3. 4b: Oblique view showing suspension points 4c: Left view - control PC, IMU, GPS 4d: Right view – camera and battery pack ROVER The autonomous rover platform used by PRoViScout is based on a robuCAR-TT chassis [8]. The rover platform, known as “Idris” (fig. 2), is a 4-wheel drive, all-terrain vehicle. On-board systems control movement and steering in both autonomous and teleoperated modes, and also implement safety features such as obstacle detection and emergency stop. The rover has a substantial payload capacity of up to 150kg and can provide accommodation, power and network services for a number of independent payload systems. 4. 4a: Bottom face with camera aperture AEROBOT Raw data for DEM generation will be provided by the tethered aerobot. This consists of a helium balloon platform capable of taking wide-angle images up to a height ceiling of 100m. It is equipped with a lightweight, high-resolution monochrome camera that Figure 4. Views of prototype aerobot hardware The following components are mounted on the framework: control computer, inertial measurement unit (IMU) and global positioning system (GPS) device (fig. 4c); high-resolution camera and battery pack (fig. 4d); miniature pulleys for suspension from the envelope (fig. 4b). The downwards-facing camera aperture can be seen in fig. 4a. The aperture is covered by an infra-red blocking filter and the camera itself is shrouded in dark material to keep out stray light. 4.2. Camera The camera used in the current version of the aerobot is a Prosilica GC2450 [9]. This camera has a Sony ICX625 monochrome CCD sensor with a resolution of 2448x2050 (5 Mpixels) and a gigabit ethernet interface. It is capable of imaging up to 15 frames per second at full sensor resolution with 8 or 12 bit pixel data. The camera allows operation modes such as single or triggered frame capture and supports configurable autoexposure and auto-gain algorithms. Image sub-framing and pixel binning are also possible. The camera is currently fitted with an 8mm focal length c-mount lens adjusted to have a hyperfocal distance of 6m; thus targets from 3m to infinity are well-focused. The optical characteristics of the camera system are summarised in tab. 1. From this table it can be seen that the camera can distinguish ground features down to a size of approx. 5cm at 100m altitude, and approx. 2.5cm at 50m altitude. Table 1. Aerobot camera optical properties Sensor Sony ICX625 monochrome Active sensor size 8.45mm x 7.07mm (2/3 fmt) Resolution 2448 x 2050 pixels Lens focal length 8mm Hyperfocal distance 6m (focus 3m – infinity) Field of view 55.65° x 47.69° Ground footprint 106m x 88m at 100m high 53m x 44m at 50m high Pixel field of view 0.0247° (0.431 milliradians) Ground pixel resolution 0.043m at 100m high 0.022m at 50m high and smoothed attitude readings are read on demand by the aerobot control computer. Internally, the sensors are read and processed at 100Hz. Positional information is provided by a miniature commercial off-the-shelf GPS unit (“dongle”) that plugs directly into a USB port and provides standard NMEA data to the aerobot control computer. After settling, this can provide latitude and longitude information to an accuracy of (at best) approximately 10m. Altitude information is also available from the GPS unit, but this has a 30m accuracy at best and so is of limited use. 4.4. Control and Interfaces The aerobot is controlled by an onboard Fit-PC2i miniature computer [11]. This device measures 101mm x 115mm x 27mm. It has a 1.6 GHz CPU, 2GB main memory, 2 x gigabit ethernet ports, 4 x USB 2.0 ports, 1 SD card slot and built-in WiFi. The computer weighs 0.37kg and consumes 6-8W of power in normal operation. We have added a fast 60GB solid state disk to store the operating system and captured data. The computer runs a version of the Linux operating system. 4.3. Position and Attitude Sensors Two additional sensor subsystems are used to provide estimates of the aerobot position and attitude. This information gives context to the captured images and is used to inform the subsequent image co-registration and DEM generation processes. A Sparkfun RAZOR 9-DOF IMU is used to provide roll and pitch angles and tilt-compensated magnetic heading [10]. This miniature IMU has 3-axis magnetic, acceleration and gyro sensors constructed using MEMS technology. It also supports custom firmware. Communication with the unit is via a serial connection, and this is implemented by a USB-to-serial piggyback board that also provides power for the IMU. Filtered Figure 5. Aerobot control system The aerobot computer handles all of the aerobot sensors and provides control and data access to them from other systems. Fig. 5 shows a block diagram of the overall system. A server process directly handles the camera and sensors and offers a command and data connection using an efficient binary protocol over TCP/IP. Using a portable client library as an adapter, this protocol can be used to operate the camera and obtain sensor data from any other network-attached system. 4.6. Envelope and Suspension The protocol is neutral about client location, but bandwidth limitations may restrict remote access. The raw data frames from the aerobot camera are some 5MB in size, and take several seconds to transfer over a good wireless network link. When reception conditions are poor, this may lengthen considerably. The envelope currently used with the aerobot is a 1.8m diameter spherical HDPE balloon, providing approx. 2kgf (approx. 20N) of lift. For the tether, 36kg (80lb) breaking strain Spectra fishing line is used in conjunction with a speed-controlled electric winch mounted on the rover platform. This tether combines high strength with low mass, and should withstand tether forces up to approximately 350N. Because of this, the usual mode of operating the aerobot is to issue high-level image sequence commands that are executed on the aerobot itself, with the images being stored in predefined locations on the solid state disk. Sample images can then be downloaded over the wireless network for examination, and the bulk data copied from the aerobot after it the imaging run has been completed. The protocol library has been implemented in C, Java and Python on several systems. For the purposes of PRoViScout integration, a CORBA interface will be implemented to enable commanding of the aerobot from other subsystems within the project. The aerobot is suspended from the balloon tether using a Picavet suspension [12] connected to both ends of a lightweight rigid rod (fig. 6). This is a technique often used in kite aerial photography. Since the connection to the balloon itself is a single thread, the platform is not affected by any rotational motion of the balloon, and the Picavet suspension keeps the platform level despite changes in the angle of the tether. The platform can still be affected by horizontal swaying movements and random air turbulence. 4.5. Power and Mass Power for all components of the aerobot is provided by a single Lithium-Polymer (Li-po) battery pack with a nominal voltage of 11.1V. In laboratory tests, with all components in use, the aerobot total current draw at 11.1V was between 900mA and 1100mA (power consumption about 10-12W). Based on these figures, a single 2200mAh battery pack should power the aerobot for up to 2 hours before needing replacement, and this estimate has been borne out by experience during field tests. To avoid the possibility of over-discharging the battery pack, aerobot flights are normally restricted to about 90 minutes duration. The total weight of the aerobot is currently 1.2kg including all components. The net maximum payload of the currently used envelope when fully inflated with fresh helium is approx. 2kg, giving a payload mass margin of 0.8kg. This is equivalent to 7.85N excess lifting force (though the weight of the tether and attachments will further reduce this). Although lightweight items have been used wherever possible, the framework itself has not been optimised with respect to weight. It may be possible to replace some of the metal panels with carbon fibre composite to reduce weight. Also, the infra-red blocking filter currently fitted is larger than necessary and could be replaced with a smaller and lighter one. However, the planned multispectral filter wheel will also add to the payload mass. This will be offset to some extent by the use of an OEM board version of the camera. Figure 6. Picavet suspension 5. FIELD TESTS 5.1. Initial Test Flight To test the envelope, suspension and basic hardware, an initial test flight was conducted during January 2011 at the Aberystwyth University robotics workshop at Llanbadarn. On a clear day with a low steady breeze, the aerobot was launched from a fixed platform and allowed to ascend until approximately 110m of tether had been paid out. Tests were made of platform stability, wireless connectivity and basic image capture. Fig. 7 shows the “first light” image from the aerobot camera. Accurate height measurements were not available, but from the apparent sizes of objects in the image, the aerobot height was estimated to be around 25m. From the initial tests, the platform stability was found to be good under static conditions, i.e. when the system was left to settle. The envelope tended to move around more during deployment and retrieval, though this was not recorded quantitatively. It was also noted that, although the wind was light, it did result in the tether finally settling at quite a shallow angle. This meant that the anchor point (notional rover position) was not visible in the images taken. Unfortunately, due to adverse weather conditions – specifically, increasingly high and gusty wind – it was found to be impossible to safely launch the aerobot on this occasion. After several attempts (and a couple of collisions), the aerobot was removed from the envelope and taken to the top of the nearby cliffs, about 15-20m above the beach level. Here a number of image sets were captured while the aerobot was pointed manually at the scene below. The image quality from the aerobot camera on this initial test was acceptable, however it was noticed that the infra-red sensitivity of the camera in conjunction with the lens being used resulted in a loss of sharpness. The lens was designed for visible light use, and was not achromatic in the infra-red. Hence the infra-red component of the image (e.g. the white appearance of the foliage) was not as sharp as the visible component. Because of this, it was decided to add an infra-red blocking filter to the aerobot camera system, at least while this particular lens was in use. Figure 8. High winds at Clarach Bay beach Although several image sets were obtained, the pointing of the aerobot was not well controlled, and this combined with the difficult terrain at the site (not all sections of cliff top were safely accessible) meant that there was little or no overlap between successive image sets. This made subsequent DEM reconstruction difficult or impossible. However, a 3D laser scan of the beach was successfully captured for later use. Figure 7. "First light" image from aerobot initial test. The sheep visible in this image are approx. 1m long. 5.2. Field Test at Clarach Bay An outdoor field test of the aerobot was scheduled for February 2011 as part of PRoViScout project. The test was to take place at Clarach Bay beach, Aberystwyth. This location has been characterised previously as a visual Mars analogue test site [6]. A panoramic stereo reconstruction of the terrain can be viewed at [13]. For this test, the aerobot was to be tethered to the rover, which was to traverse a path along the beach, stopping at regular intervals to take a series of aerobot images. The rover was also to take a 3D laser scan of the terrain at each waypoint as a “ground truth” measurement for later comparison with the generated DEM. Figure 9. Aerobot image of Clarach Bay beach from the cliff top (~18m) 5.3. Controlled Test at Robotics Workshop In place of the planned field test at Clarach Bay, a controlled test of aerobot image capture was conducted at the robotics workshop at Aberystwyth University. A test area was prepared with a number of imaging targets scattered over an area of sloping terrain. Some obstacles were placed within the test area, while several other natural features were included. • An image is captured at the nominal shutter time and also at factors of 0.5 and 1.5 times the nominal shutter time (exposure bracketing). This whole sequence is repeated a number of times at each viewpoint (10 in initial trials, later reduced to 4). The combination of exposure bracketing and multiple image capture is intended to allow for movement of the aerobot platform and rapidly-changing light conditions. The aerobot was disconnected from its envelope and tether, and was instead mounted on a cantilevered platform overlooking the test area. The height of the platform was 4 – 5m above the ground level. Figure 11. Aerobot image capture with overlap Figure 10. Captured aerobot images Images were captured by the aerobot from a number of overlapping viewpoints above the test area, moving around a corner and back again. Altogether, images were captured from 23 different viewpoints (a total of 690 images or 3.3GB of data). Representative captured images are shown in fig. 10. The data captured during this test was successfully used for DEM reconstruction. The methodology used for this process is outlined in section 6. 6. The Aerobot images (fig. 10) contain rough orientation parameters (position from GPS and pointing from IMU). Together with the disparities obtained from automatic interest point matching (shown in fig. 12), they are fed into a bundle block adjustment procedure to produce refined camera position, orientation and interest point 3D coordinates (fig. 13)[4]. VISION DATA PROCESSING The aerobot is tethered to the rover platform, thus it is possible to take images from multiple viewpoints with overlapping footprints (fig. 11). Images from the aerobot are rectified and co-registered to produce a consistent aerial view of the rover mission environment. At each viewpoint, images are captured by the following sequence: • The camera auto-exposure algorithm is run once, to set initial shutter time. Target mean pixel value is set to 40% of maximum. • The computed shutter time is read from the camera (nominal shutter time). Figure 12. Interest point matching between two images A following 3D reconstruction uses disparities and image orientations for dense DEM generation (figs. 14 & 15). The DEM is used in PRoViScout as an initial map for scientific context and navigation [5]. In addition, overlaying multispectral views of the same terrain on the DEM will allow for a spectral and therefore mineralogical analysis of the area, further informing the initial mission plan. Figure 13. Left: Camera x-y positions; Right: Camera pose and interest points recovered in 3D. Images © Center for Machine Perception, Czech Technical University in Prague. Figure 14. Generated DEM as elevation map • The physical construction of the aerobot framework needs some attention, to reduce weight and further increase robustness in case of collisions. • The limited stability and lifting power of the envelope used has presented problems during periods of strong wind. In order to broaden the range of weather conditions under which the aerobot will be flown, alternative envelope configurations should be considered. A larger sphere would increase lift force and hence improve tether angle, while a more “blimp-like” envelope would allow increased stability at higher wind speeds. With both of these options there is a tradeoff of lifting power against the cost of helium to fill the envelope (which can be quite substantial). • The current altitude measurement is inaccurate. This is not a problem for DEM generation provided that an object of known scale (for example, the rover) appears in the image sequence. If required, a more accurate altitude estimate could be obtained from a suitably calibrated barometric altimeter, or by measuring the tether angle and length with suitable instrumentation. A major outstanding item of work is the provision of a filter wheel for the aerobot camera. Carrying a selection of broadband (visual red-green-blue) and narrowband (geology) filters, this will allow not only colour image reconstruction for human consumption, but also spectral analysis of the terrain and its classification according to detected mineralogy. As an example of the intended output, fig. 16 shows a DEM with overlaid mineralogy information produced by the CRISM instrument of NASA's Mars Reconnaissance Orbiter. Figure 15. Generated DEM rendered in 3D 7. DISCUSSION AND FUTURE WORK The design of the aerobot is still at an early stage, though the results from the initial tests are encouraging. We have shown that the camera system and software pipeline is capable of producing a valid DEM from captured images. The aerobot control systems and interfaces are also fully functional. Nevertheless, several points present themselves for further consideration: Figure 16. Mineralogy map of Nili Fossae region, Mars Image: NASA/JPL/JHUAPL/University of Arizona/Brown University ACKNOWLEDGEMENTS WEB REFERENCES The research leading to these results has received funding from The European Community's Seventh Framework Programme (FP7/2007-2013), Grant Agreement No. 241523 PRoViScout. Retrieved on 2011-04-01 Some images were provided by project partner Czech Technical University, Prague. REFERENCES [1] Paar, G., Woods, M. & the PRoViScout Team. FP7SPACE PRoViScout – Planetary Robotics Vision Scout. Proc. 11th Symposium on Advanced Space Technologies in Robotics and Automation, ESA/ESTEC, The Netherlands, 12-14 April, 2011 [2] Dave Barnes, Andy Shaw, Phil Summers, Roger Ward, Mark Woods, Malcolm Evans, Gerhard Paar, Mark Sims. Autonomous Image Based Localisation For A Martian Aerobot. ISPRS Symposium “From Sensors to Imagery”, Paris, July 3-6, 2006, 2006 [3] Honary, E., McQuade, F., Ward, R., Woodrow, I., Shaw, A., Barnes, D., Fyfe, M. Robotic experiments with cooperative aerobots and underwater swarms, Robotica, Vol. 27, Issue 01, pp 37-49, 2009. [4] Jancosek, M. and Pajdla, T. Hallucination-free Multi-View Stereo, RMLE 2010 (ECCV workshop), Hersonissos, Heraklion, Crete, Grece, September 5–11, 2010 [5] Woods, M., Shaw, A., Rendell, P., Long, D., Paar, G. High-Level Autonomy and Image prioritisation for long distance Mars Rovers. Proc. 11th Symposium on Advanced Space Technologies in Robotics and Automation, ESA/ESTEC, The Netherlands, 12-14 April, 2011 [6] Pullan. D. Robotic Geology Assessment of Clarach Bay, Aberystwyth, UK. Technical Reference, PRoViScout Project, July 2010 [7] PRoViScout official project web page: http://www.proviscout.eu [8] robuCAR-TT mobile platform: http://www.robosoft.fr/eng/sous_categorie.php? id=1025 [9] Prosilica GC2450 GigE camera: http://www.alliedvisiontec.com/emea/products/cam eras/gigabit-ethernet/prosilica-gc/gc2450.html [10] Sparkfun RAZOR 9-DOF miniature IMU: http://www.sparkfun.com/products/9623 [11] Fit-PC2i miniature computer: http://www.fit-pc.com/web/ [12] Picavet suspension system: http://en.wikipedia.org/wiki/Kite_aerial_photograp hy#Picavet_suspension [13] Rendered video fly-through of Clarach Bay stereo panorama reconstruction: http://www.youtube.com/watch?v=6gRo8QSXX5c