Stanford's Stanley Wins Driverless Race
These days the winner of an auto race usually does smoky burnouts on the track and then climbs the chain-link fence to salute the crowd. But the recent Defense Advanced Research Projects Agency (DARPA) Challenge wasn't like that, because the winning car didn't have a driver. In fact none of the competing vehicles had a driver because that's what the federal agency was looking for -- a vehicle that could complete its 132-mile desert course with no driver aboard.
Last year Carnegie Mellon University's Sandstorm was the most successful robot-driven vehicle when it traversed less than eight miles of a 170-mile course. This time the Stanford Racing Team's autonomous robotic car, Stanley, equipped with serious amounts of artificial intelligence, negotiated the somewhat shorter off-road course south of Las Vegas in a little less than seven hours. The winning finish qualified the Stanford team for DARPA's $2 million race prize payout and captured it a lofty place in the history of robotics and technology.
"The impossible has been achieved," said team leader Sebastian Thrun, an associate professor of computer science and director of the Stanford Artificial Intelligence Laboratory, with more than a trace of hyperbole. (Certainly at an institution of higher learning as prestigious as Stanford they teach that if something can be achieved it is not "impossible.")
A modified Volkswagen Touareg, Stanley earned the prize and the glory by completing the rugged off-road course with the quickest time, 6:53:58. Only four other cars of the 23 in final contention managed to finish. Two cars from Carnegie Mellon University, H1ghlander Hummer and Sandstorm, followed Stanley with times of 7:04:50 and 7:14:00 respectively. The vehicle entered by a team hailing from Metairie, Louisiana, finished in 7:30:16. (Some of its team members had lost their homes to Hurricane Katrina but soldiered on in the DARPA Challenge anyway.) The last vehicle to finish, a yellow behemoth of a military truck named Terramax, didn't cross the line until nearly a day after the other vehicles finished. Its human-piloted chase vehicle broke down en route so DARPA officials had it stop repeatedly during its run on the course.
Perhaps because DARPA is a government agency, it didn't declare a winner until some 24 hours after the Stanford entrant had crossed the finish line. But when the belated news came the Stanford team went crazy -- writing algorithms and testing hypotheses late into the afternoon. In a fit of exuberance team members also poured two enormous Red Bull cans containing ice water over Thrun's head and shortly after, Thrun and postdoctoral researcher Mike Montemerlo were hoisted aloft to ride on the shoulders of their sleep-deprived teammates.
"We had a good day," Thrun said playfully after a throng of media perched on a nearby platform pressed him to say something.
How did Stanley manage to set the pace? Just the same way you get to Carnegie Hall -- practice, practice, practice. Stanley "learned" during countless hours of desert testing in the months leading up to the race, often going days at a time without a drink of water or a cigarette. Equipped with a wide variety of sensors and a heap of custom-written software including machine-learning algorithms, Stanley grew smarter with practice. Eventually the special Touareg became a master of finding the path, detecting obstacles and avoiding them while staying on course.
The defense agency sponsored the competition because applications of the technology could range from the development of unmanned ground vehicles for dangerous military missions to driver-assistance systems that keep civilian drivers, passengers and pedestrians safe.
Based in Villeperce, France, Driving Today Contributing Editor Tom Ripley writes about the auto industry, the human condition and, in this instance, the inhuman condition.