DARPA wants to know how they got as far as they did
Once the robot vehicles took off, no remote-control operations were allowed.
Courtesy of defense Advanced Research Projects Agency
Although no contestant reached the eight-mile marker in the Defense Advanced Research Projects Agency's first self-navigating vehicle race, DARPA still judged the recent event a success.
Several of the teams are now reorganizing for the next Grand Challenge, probably next year. DARPA has a four-year permit from the Bureau of Land Management to use selected tracts of land in the Mojave Desert for the race.
'It didn't matter to us if anybody completed the course' this year, DARPA director Anthony Tether said. What the agency really wanted was for more scientists, engineers and hackers to think harder about how to write navigation software.
'We're going around to all the teams that participated to find out what went through their minds,' Tether said. The agency would award small contracts for sharing the innovations, he said.
Staging the race this year cost the agency about $13 million, less the $1 million purse that went unclaimed. Tether estimated DARPA was getting back at least four or five times its investment from free research by teams trying to build competitive advantages into their vehicles.
One by one, 13 vehicles prequalified by DARPA set off from Barstow, Calif., on March 13 to find their own way 142 miles across the desert to the finish line in Primm, Nev.
What made the event different from the usual off-road races was that once the robotic vehicles were turned loose, they were entirely on their own. No remote-controlled operations were allowed. A manned control vehicle did, however, follow each contestant for emergency stops.
Autonomous vehicles are nothing new; robotic floor cleaners, lawn mowers and farm tractors are commercially available. But none so far can navigate through rock-strewn territory with steep slopes, dry lake beds, mud, cattle guards and, of course, lots of sand.
Congress has set a goal for the Defense Department to make one-third of all its ground vehicles unmanned by 2015. DARPA researchers have been trying to crack the machine cognition problem for years, with limited success.
The race's vehicles were mostly based on off-road favorites such as Hummers, four-wheel-drive pickups and all-terrain vehicles. The various teams rigged commercial actuators, hydraulic pumps and servers to steer, accelerate and brake.
About three hours in advance, DARPA issued each team a CD-ROM with the latitude and longitude of about 2,000 checkpoints and speed limits on the race route. Commercial Global Positioning System receivers provided coordinates to the vehicles. Off-the-shelf radar furnished long-range vision, while stereo cameras and light-detection and ranging devices known as lidar gave more detailed information about closer objects.
Most teams used commercial computers or notebook PCs running Linux or a real-time operating system, though some built in rugged notebooks or shock-resistant hard drives.
The component that most interested DARPA, however, was the software that would process the sensor information and make decisions on how to move forward.
'The secret sauce is in interpreting the sensors and figuring out what to make the vehicle do,' Tether said. 'Not only to be able to say, 'Yes, that is the path over there,' but to transform that decision into going faster, putting a brake on, turning left.'
More than 80 teams applied after the announcement in early 2003, though DARPA whittled that number down to 25 finalists. The teams ranged from highly organized consortia of companies and academics to Southern California gearhead shops specializing in high-performance racing automobiles. A few high school and university teams participated.Tortoise not in race
Before the race, DARPA assigned 20 biologists to chart desert tortoise burrows along the route and place protective barriers around them them.
DARPA staggered the starts at five-minute intervals and arranged the lineup based on each vehicle's chances of completing the race. The strongest contenders got to depart first.
The first vehicle, a Hummer customized by Carnegie Mellon University's Red Team, represented the largest investment: an estimated $3 million worth of parts and labor. The vehicle took off smoothly from the starting gate and made its way down a dirt road and around a series of curves that doubled back behind the staging area.
Other launches were less smooth. Team Caltech's white sport utility vehicle lurched tentatively out of the starting gate and stopped every 30 seconds or so to recalculate its course.
Team CajunBot's all-terrain vehicle ran about 20 yards before sideswiping a concrete bunker. It tried to extricate itself but rolled over a bush.
Team ENSCO's six-wheeled vehicle speedily left the starting gate and took off down the straightaway, but it soon jumped into bushes alongside the road and flipped over.
After a few hours, even the best vehicles had been judged disabled. Carnegie Mellon's Hummer drove 7.4 miles before dying a smoky death in 'what is known in robotics as a system failure,' Red Team technical adviser Jay Gowdy said. A series of minor failures cascaded into total inoperability.
The chain of problems started when the Hummer broke off its front fender, losing its sensors and leaving only the less detailed GPS coordinates for navigation. The Hummer soon veered off course, tearing up its underside and eventually getting wedged on a large rock.
As it tried to break free, Gowdy said, the rubber torn from its tires landed near the engine and caught fire.
Although no vehicle traveled more than a twentieth of the course, the performances were 'extremely impressive,' said Thomas Strat, DARPA deputy program manager for the event. He noted that after the first five miles, the course got considerably more difficult with steep, narrow slopes and a switchback.