|
Team Overbot is a group of enthusiasts, engineers, and partners building an autonomous robot vehicle for the October 8, 2005 DARPA Grand Challenge. We're talking a trek close to 150 miles through all kinds of terrain. And at the end of the road is a $2,000,000 prize.
Last year, the course included a 142-mile trek from Barstow, CA to Primm, NV over well-traveled utility roads, switch-backs, severe elevation changes, blind turns, and sheer drops. None of the vehicles finished the course. Team Overbot dropped out before the race last year, but this year, they're ready.
The Contest
The DARPA (Defense Advanced Research Projects Agency) Grand Challenge was designed as a field test and as a tribute to the U.S heritage of innovation, risk-taking, and a sense of team spirit. It brings together individuals and organizations from industry, the R&D community, government, the armed services, academia, students, backyard inventors, and automotive enthusiasts in the pursuit of a great technological challenge. The competition's aim is also to accelerate R&D in autonomous ground vehicles that can be used to save lives on future battlefields. DARPA expects to qualify 20 vehicles to participate out of the more than 100 applicants.
Design Methodology
The Overbot vehicle is a commercial 6-wheel-drive all terrain Polaris Ranger Series 11 all-terrain vehicle manufactured by Polaris Industries (Minneapolis, MN). The front four wheels are on independent swing arm suspensions, and the rear axle is rigid, but on a swing assembly. The overall vehicle, fully-fueled, weighs approximately 1,900 pounds, with a maximum range of 350 miles.
The front two wheels are steerable, four of the wheels are equipped with hydraulic brakes, and all wheels are driven when in 6WD mode. An electrical/vacuum system switches the vehicle from 2WD to 6WD. The vehicle uses servomotors to drive the steering, brake, transmission, and throttle. The engine choke is actuated using a solenoid. The laser range-finder atop the vehicle is tilted via a servomotor as well.
The Overbot vehicle computing systems include three small industrial control microcomputers for sensor and actuator control. Two larger computers are used for vision and navigation processing. All the computers use an IA-32 architecture and are interconnected via a 100baseT networking system.
Team Overbot developers chose a navigation approach that first measures terrain and obstacles using a variety of sensors. From the data collected, the system builds a local terrain map of the immediate vicinity of the vehicle. This map is probabilistic and contains uncertainty information. Next, an attractive/repulsive field type planner generates trajectories. In addition, a visual "road follower" attempts to recognize road surfaces and add them to the vicinity map, so that if a road is present and going in the desired direction, it will be used.
Out on the Challenge course, the vehicle is designed to head toward the next GPS waypoint, using a higher level processor, unless it has encountered an obstacle. According to John Nagle, project manager for Overbot, "Escaping from local minima is the job of the higher level processing system. Internally, we call this the backseat driver, because it has no direct authority over the control system." This processor will then mark untraversable areas in its vicinity map and attempt to work around them, directing the Overbot to back up if necessary.
At a lower level, processing algorithms for individual sensors exert veto power over high-level decisions, which may result in the vehicle stopping suddenly if an obstacle is detected. "Our general approach is not to out-drive our stopping distance. We insist on good ground profiling data from the laser rangefinder out to our stopping distance. Pitch will be factored into the stopping distance computation, and rough ground will be covered at slower speed so that the vehicle sees shock levels well under 1G vertically. We will not exceed 40 mph at any time," Nagle says.
|
