In a race that began and ended in a casino parking lot and traversed132 miles of desert southwest of Las Vegas on Oct. 8, the StanfordRacing Team's autonomous robotic car, Stanley, won big. Theartificially intelligent car traversed the off-road course in a littleless than seven hours, yielding both a $2 million payout and a loftyplace in the history of robotics and technology.
“The impossible has been achieved,” said team leader SebastianThrun, an associate professor of computer science and director of theStanford Artificial Intelligence Laboratory. Simply by finishing thecourse, Stanley and four other cars showed that machines can be made todrive safely and speedily over rugged terrain without any human help.
Stanley earned the prize and the glory in a contest sponsored by theDefense Advanced Research Projects Agency (DARPA) because the modifiedVolkswagen Touareg completed the rugged, off-road course with thequickest time, 6:53:58. Only four other cars of the 23 in contentionmanaged to finish. Two cars from Carnegie Mellon University followedStanley with times of 7:04:50 and 7:14:00. In a poignant victory, ateam hailing from Metairie, La., finished in 7:30:16. Some team membershad lost their homes to Hurricane Katrina and the entire team lostprecious practice time because of the storm.
The last vehicle to finish—a yellow behemoth of a military trucknamed Terramax—did not do so until the morning of Oct. 9. DARPAofficials forced it to pause repeatedly during its run, and eventuallyovernight, in part because of the breakdown of its human-driven “chase”vehicle. Because of the overnight delay, DARPA ultimately could notprovide official times for all participants and declare Stanley thewinner until almost 24 hours after it crossed the finish line.
Thrilled with Stanley's performance, the Stanford team membersdidn't seem glum about having to wait. In addition to dozens ofStanford faculty, students and staff from the School of Engineering,the team included a large contingent of engineers from supporterVolkswagen of America's Electronics Research Lab. Other key supportersincluded the Palo Alto venture capital firm MDV—Mohr Davidow Ventures,energy drink maker Red Bull and tech startup Android.
The thrill of the race
Without knowing whether it was the overall winner, the Stanford teamSaturday still had the unique joy of being the first to celebratefinishing the course, and the group took full advantage. Stanleycruised across the finish line early in the afternoon. As DARPADirector Tony Tether waved the checkered flag, the team and the packedgrandstand roared. Team members poured two enormous Red Bull canscontaining ice water over Thrun's head and shortly after, Thrun andpostdoctoral researcher Mike Montemerlo were hoisted aloft in asleep-deprived frenzy of jubilance and relief and riding on theshoulders of their teammates.
“We had a good day,” Thrun said playfully after a throng of mediaperched on a nearby platform pressed him to say something. In shortorder he was bobbing in a sea of interviewers, with fuzzy boommicrophones floating over his head like a swarm of flying squirrels.
Stanley was the second vehicle to leave the starting line, fiveminutes after Carnegie Mellon University's H1ghlander Hummer and fiveminutes before that university's second entry, Sandstorm. Sandstorm wasthe most successful robot during the first Grand Challenge last year,when it traversed less than 8 miles of a 170-mile course.
For most of the race Stanley trailed CMU's hulking red H1ghlanderuntil just past the 100-mile mark, when Stanley overtook the laggingHummer. By erasing CMU's head start, Stanley vaulted into front-runnerstatus.
It was not much later that Stanley wowed the crowd again when itheaded into Beer Bottle Pass, a treacherous and windy mountain passwith sharp turns and sharper cliffs along one side of the road. Via aDARPA live video feed from the entrance to the pass, spectators couldsee Stanley drive into view, rounding the corner smoothly, as if therobot had not only artificial intelligence but real confidence.
Smart brain, strong body
Stanley got that smart by learning during countless hours of deserttesting in the months leading up to the race. Equipped with a widevariety of sensors and a heap of custom-written software includingmachine learning algorithms, Stanley grew smarter with practice.Eventually it became a master of finding the path, detecting obstaclesand avoiding them while staying on course.
Applications of the technology could range from the development ofunmanned ground vehicles for dangerous military missions to driverassistance systems that keep civilian drivers, passengers andpedestrians safe.
Those applications, however, will have to wait. After a grueling 18months of development, culminating in a 4 a.m. wake-up call to startSaturday's race, Thrun said, “The next challenge is to get some sleep.”
Cite This Page: