Demonstrating unprecedented prowess in robotics and artificial intelligence, five unmanned, autonomous vehicles last weekend dashed all the way across a rugged, 210-kilometer stretch of the Mojave Desert.
Eighteen other vehicles also began the off-road race near the California-Nevada border on Saturday morning, Oct. 8, but fell victim to such difficulties as mechanical breakdowns, sensor failures, and computer glitches.
The successful racers have “proved that it’s possible to develop truly autonomous ground vehicles that move at militarily relevant speeds,” says Ron Kurjanowicz of the Arlington, Va.–based Defense Advanced Research Projects Agency (DARPA), which sponsored the race.
Last year, none of 15 vehicles in the first DARPA-sponsored Grand Challenge race made it to the end of a different Mojave Desert route.
DARPA has promoted the races in order to accelerate development of autonomous ground vehicles for military use. In 2000, Congress mandated that one-third of future army vehicles should be autonomous.
Eventually, robotic trucks using technology showcased in the race, such as laser-based obstacle sensors, might deliver supplies to troops without risking the lives of human drivers, says Kurjanowicz, who manages the Grand Challenge program. Whether robotic fighting vehicles might someday participate in combat is an open question, he adds.
The winning vehicle was a blue 2004 Volkswagen Touareg sports utility vehicle (SUV), nicknamed Stanley and outfitted with five laser-based sensors peering down from its roof. It was created by researchers at Stanford University and their industry partners, including Volkswagen.
To be considered for the $2 million prize, contenders had to complete the course within 10 hours. Stanley buzzed to the finish line in just under 6 hours, 54 minutes. The spunky racer averaged a little over 30 kilometers per hour (kph), with peak speeds exceeding 60 kph.
The hodgepodge of other finishers included military-style vehicles, a Ford Escape hybrid SUV, and a lumbering, 16-ton truck.
Much of the difference in the outcome of this year’s race and last year’s stems from improvements in software and sensors, says Sven Strohband of Volkswagen of America, who led the car company’s efforts on Stanley.
Those advances include a refinement of the Stanford vehicle’s obstacle-avoidance system. To train the vehicle’s autonomous systems to react more as people do, the team relied on data collected as human drivers of the vehicle negotiated various types of terrain, says Michael S. Montemerlo, who headed the team’s software development. The resulting cut in the rate of false alerts that an obstacle was present—from about 1 in 8 to 1 in 50,000—”was a major breakthrough for us,” Montemerlo says.
William “Red” Whittaker, leader of the second– and third-place teams, both from Carnegie Mellon University in Pittsburgh, says that he expects some of the racers’ technologies to eventually alter the average person’s daily drives. Other commercial impacts are anticipated in trucking, construction, and agriculture.
