On 7 September 1997, the IROS Planetary Rover Workshop was held in Grenoble, France as part of the 10th International Conference on Intelligent Robots and Systems. Seven presentations were made from people representing organizations in Europe and the USA:Although all institutions had vehicle prototypes, the emphasis of the presentations varied greatly, focusing on results at many levels: In the notes that follow, I have attempted to provide some feedback on how the workshop presentations went. I did not attempt to include specific details about the vehicles here, since that information is already present in the workshop slides themselves. I realize that make it difficult for a general audience to compare them, but I have provided pointers to any online information I have found or been told about. If you are really interested in the workshop slides, I suggest you contact the organizer, Dr. Richard Volpe of NASA JPL.
CNES developed the EVE rover using VNII Transmash's Marsokhod platform with a stereo head and inertial sensors.
Speed: 10-15 cm/sec
The vehicle has 12 degrees of freedom in the chassis.
Onboard sensors include stereo vision cameras with 70 degree FOV mounted on a pan/tilt head, inclinometers, and wheel encoders.
EVE is used primarily as platform for low data-rate Lunar teleop and Martian autonomy studies.
The first talk, presented by a CNES representative, emphasized varying mission data rate scenarios. The rover capabilities were changed to compensate for each of the mission scenarios (e.g., increased autonomy for low bandwidth Mars operations, more user interaction for high bandwidth Lunar operations).
Field testing occurs in their Moonyard (50 m x 70 m) and Marsyard (20 m x 40 m).
The second talk (by a LAAS-CNRS representative) presented results of the autonomous system on the EVE robot, mentioning their plan to use a future robot to explore Antarctica looking for meteorites.
Environment modeling is the key component of autonomy. They use stereo vision for geometric analysis, video texture for remote terrain classification, and perform self-localization using terrain peaks as features.
Long term research intent is to use spherical attribute image to blend 3D info, luminance, and texture info onboard.
EVE reacts to dangerous terrain using internal attitude sensors; when the vehicle starts pitching too much, these sensors cause it to react and move to a safer position.
CNES is beginning development of EVE's successor, called IARES, which will have much greater mobility capabilities with an enhanced Marsokhod chassis.
IARES == EVE++ . It will have additional power, and the vehicle chassis has 7 additional degrees of freedom (compared with EVE).
They plan to improve navigation by increasing processor power, reducing map and image resolution, and use better navigation algorithms (developed by LAAS).
IARES has many new mobility features: wheel walking, canyon crossing, cross-slope driving.
A 3D terrain and vehicle simulation is already functional.
The new mobility features are quite interesting, they do appear to solve interesting traversability problems.
The chassis is currently built, but has no sensing onboard, and no autonomous capability for invoking the new mobility features.
The questions asked seemed to implicitly criticize their choice of greater mobility at this time; additional power, weight, and complexity seemed unjustified (but the answer was that they want to study what can be gained by these new capabilities).
DLR is developing a small tethered explorer for Lunar operations under ESA's Euromoon 2000 initiative: it will seek out volatiles at the Lunar south pole.
The Mobile Instrument Deployment Device (MIDD) is to be used in conjunction with a longer range rover (e.g., something like the EVE vehicle). The MIDD will have a high science payload/mass ratio, meaning the emphasis will be on science instruments rather than long distance traverse capability.
There is a definite emphasis on instruments at the expense of mobility: the 5kg rover will be able to handle 10cm step obstacles and travel at 1mm/sec.
MIDD is designed for maximum 50 day lifetime.
The presentation outlined some very methodical component testing: vacuum chamber with well-characterized dust particles, linear rail with dust/dirt and force sensing, tether deployment force sensing. See the DLR Planetary Simulation Facility web page for more info.
The front wheel assembly includes rotating dust seals (annular metal springs that exert axial force).
Only 20% gear train efficiency because of worm gears on test article. Gear ratio 1000:1.
Current design uses brushless motors, why use them for flight (Sojourner has brush motors)? The answer is that they are studying more than just ESA's 2003 Mars Express mission.
Very simple tether deployment; drive away and it unravels (still in a loop).
The complete vehicle is expected by the end of 1998.
MIT's Pebbles robot uses behavior-based reactive control; no high level terrain or world modeling. A special purpose "cheap vision machine" provides high throughput image processing that enables color edge discrimination in real time (4 Hz image updates from a single camera).
Pebbles' speed is at least 40 cm/sec (judging by the video footage shown; could be higher).
The vehicle has been demonstrated in MIT and JPL Mars yards.
Pebbles produces interesting low level obstacle avoidance behaviors, but this lack of 3D information means it can get confused.
Pebbles includes a compliant manipulator as well as vision system.
Core idea is: send several rovers, don't worry if they are not precise or get lost; others will survive.
Taken even further, 10-15 gram "Hopette" rovers have been studied (they look like weebles; egg-shaped objects with weights on one end). An egg-shaped container uses a mechanical spring to induce trajectories, and its payload can be repositioned to enable "aiming" of the hop. There is no need for accurate navigation, just stumble around and use RF comm to relay any interesting images. The team has demonstrated small enough CPUs and locomotion, but not yet imaging and communications
The primary focus of the talk was on the Lavic Lake field trials.
The rover performed well in desert extremes, and students from six schools drove it over the Internet (including some international schools).
Also mentioned other rover efforts at JPL, with emphasis on the nanorover.
This was the first presentation of results from the Atacama Desert Trek, a six week field test with the rover driving over 200 km in Northern Chile while controlled at remote stations thousands of miles away.
Attendees had a special interest in the autonomy results, antenna pointing technology, and public outreach.
Chilean support was outstanding.
All objectives were met.
The latest results (as of 7 September) from Sojourner were presented in read-only mode.
One interesting tidbit: loss of battery meant the onboard computer's clock was useless, hence internal safeguarding got confused. For example, there was some code that controlled a heater on the modem, and that code had different behavior depending on the time of day. In one mode it assumed if there was no lander communication then the modem must be too cold, and it would turn on the heater. But at the start of a new day the rover would typically start before the lander, which had the potential of overheating and therefore damaging the modem. This was, of course, fixed.
Sojourner has run autonomously for several short distances, using five fixed laser emitters to detect obstacles.
Sojourner has a 100 Kips flight-qualified 8085 processor.
The mission has been a more-than-complete success.