October 7, 2005 4:00 AM PDT

Robots take the wheel in desert race

Fittingly for a car race to be held in the desert outside Las Vegas, 23 teams battled long odds to make it into a most unusual competition.

This is no ordinary race. The competitors are artificially intelligent robots designed to drive autonomously, and they're facing tough terrain: A 150-mile desert course with mountain switchbacks, gullies, dry lake beds, tunnels and manmade obstacles. And the computer scientists who developed the robot racers have to balance care with speed; the robots must finish the course in under 10 hours.

"Everyone has to place a bet down on speed," said Sebastian Thrun, a computer scientist and director of Stanford University's artificial-intelligence laboratory.

Stanford Racing Team's "Stanley," a modified Volkswagen Touareg V5, is one of the finalists announced Thursday by the Defense Advanced Research Projects Agency (DARPA), the research and development unit of the U.S. Department of Defense that sponsors the desert race, which takes place Saturday.

Carnegie Mellon University's entrant, called Sandstorm, was last year's best-performing robot. This year, CMU has two cars competing. H1lander, a modified Hummer that will hold the pole position in the Saturday race, and Sandstorm, which will leave in third position 10 minutes later.

Truth is, the odds are stacked against all of them. Last year, CMU's Red Team, the best finisher, made it across only 7.3 miles of the 144-mile course in the Mojave Desert before burning out. No one claimed the $1 million prize.

"In 2004, we thought it was quite an achievement that a robot was able to go about seven and a half miles," DARPA director Tony Tether said. "But the results of this (year's semifinals) tell me that we will leave that in the dust of the Mojave."

This year, the prize is $2 million, and researchers hope their robot racers will fare better. The teams just finished eight days' worth of semifinals at the California Speedway in Fontana, Calif. Forty-three robots took turns driving 2.2- to 2.7-mile courses designed to resemble desert conditions. The semifinalists were chosen from among 195 original applicants.

Stanley meets the Challenge

The semifinal course wasn't easy. It included a 100-foot-long tunnel, more than 100 cones, gates, parked cars and piles of tires. One motorcycle robot crashed into a metal obstacle, but was able to get itself back up onto the course with the use of metal legs and carefully timed speed. Another robot ran over hay bails and ignited them. DARPA, with the use of a remote control that can pause or disable a vehicle, stopped the car and pulled it from the race.

According to several competitors, the course was altered over the eight days to prevent memorization by the robots and ensure it was more difficult each day.

Several teams finished the courses, including Stanley, Terra Max and CMU's Hummers.

Just two weeks before the semifinals, H1lander rolled over on a practice run. But during the semifinals, things went smoothly except for running over a few obstacles.

"No one can yet say what it takes to succeed in the Grand Challenge since the race hasn't occurred and the challenge hasn't been met," said professor Red Whittaker, Red Team leader, Carnegie Mellon Robotics.

The robots can drive themselves with the use of radar, vision and laser sensors fastened to the cars that can act as early warning systems, detecting close and far-range obstacles. They also draw on GPS (Global Positioning System) sensors to trace many steps.

In Stanley's case, Thrun and a team of computer scientists wrote more than 100,000 lines of code to tell it what to do. A map tells the car where to drive; a planning tool points out unsafe terrain; and a controller translates all of that into action. The software runs on six Pentium M processors, Intel-made, low-power chips originally designed for the telecommunications industry.

The race was devised by DARPA when the government asked it to produce a fleet of autonomously driven vehicles by 2015. The Grand Challenge calls on the expertise of computer scientists--whether in private business or academia--to develop the high-tech vehicles and race for the prize. DARPA hopes to use the technologies for vehicles that could one day drive onto a battlefield without endangering soldiers.

The racers won't know the exact details or location of the course until two hours before the race in Primm Valley, Nev., which begins at 6:30 a.m PDT. That way, contestants won't be able to test the course beforehand and prime their vehicles.

Among the other teams racing in the second annual DARPA Grand Challenge are Team CalTech of Pasadena, Calif.; Team Cornell from Ithaca, N.Y.; Desert Buckeyes from Ohio State University; Mojavaton from Grand Junction, Colo.; MonsterMoto from Cedar Park, Texas; Team Terra Max from Oshkosh, Wis.; Intelligent Vehicle Safety Systems I from Littleton, Colo.; and Virginia Tech Team Rocky from Blacksburg, Va.

Also competing in the race are Axion Racing, from Westlake Village, Calif.; Team Cajunbot, from Lafayette, La.; CIMAR, from Gainesville, Fla.; Team DAD, from Morgan Hill, Calif.; Team ENSCO, from Springfield, Va.; the Golem Group/UCLA, from Los Angeles; the Gray Team, from Metairie, La.; Insight Racing, from Cary, N.C; Mitre Meteorites, from McLean, Va.; Princeton University; SciAutonics/Auburn Engineering, of Thousand Oaks, Calif., and the Virginia Tech Grand Challenge Team.

The challenge "is a truly powerful mix of American ingenuity, team spirit, competitiveness, entrepreneurship, engineering and computer science," said Ron Kurjanowicz, Grand Challenge program manager.

See more CNET content tagged:
robot, course, obstacle, race, Hummer

1 comment

Join the conversation!
Add your comment
my subject
This is a comment place..
Posted by (1 comment )
Reply Link Flag
 

Join the conversation

Add your comment

The posting of advertisements, profanity, or personal attacks is prohibited. Click here to review our Terms of Use.

What's Hot

Discussions

Shared

RSS Feeds

Add headlines from CNET News to your homepage or feedreader.