Robots to race in suburbia
Posted: Thursday, August 09, 2007 6:29 PM by Alan Boyle
Today the Pentagon revealed its list of 36 contestants for the next multimillion-dollar race for autonomous vehicles - and also revealed where the robo-finalists will face off in November. It may be called the "DARPA Urban Challenge," but the race course is actually a slice of faded suburbia in the California desert.
The Defense Advanced Research Projects Agency says the competition will take place on the decommissioned George Air Force Base near Victorville, Calif. - home to a 1,000-building complex that's been used for urban combat training and has thus been dubbed "Al-George."
Despite the nickname, Al-George looks less like downtown Baghdad and more like a suburban neighborhood, complete with traffic circles and street signs. There's even a golf course nearby. Once the Army finishes its current training cycle, DARPA will clean up the site (for example, removing downed trees) and get it ready for the robots.
The qualifying event for the 36 semifinalists will be conducted from Oct. 26 to 31, and the top 20 teams will move on to the Nov. 3 finals. Top prize is $2 million, with $1 million going to the runner-up and $500,000 to the third-place finisher.
DARPA is funding the competition to promote technologies that can eventually produce smarter robotic vehicles for military applications - yes, even in downtown Baghdad. Congress has told the Pentagon it wants a third of America's ground combat vehicles controlled autonomously by 2015. Along the way, civilians will likely benefit from smarter cars and trucks as well.
During the finals, the autonomous vehicles will have to complete a roughly 60-mile course in six hours, simulating a military supply mission. Al-George was selected as the site for the competition because its street network simulates the kind of terrain U.S. forces face during overseas deployments, DARPA said in today's news release.
"This adds many of the elements these vehicles would face in operational environments," DARPA director Tony Tether explained.
But it's not enough just to hit every waypoint on the route. The robots will have to obey the rules of the road - including dealing with the traffic circles, four-way stops and merging traffic. That's a change from DARPA's $2 million Grand Challenge in 2005, which merely called for autonomous vehicles to drive over 132 miles of desert roads.
"The vehicles must perform as well as someone with a California driver's license," Tether said.
During the finals, penalties will be added to the times of the vehicles involved in traffic infractions - and the winner will be decided on the basis of the adjusted time. Thus, it's possible for a reckless Bender to finish the race first but still lose out to a safe-driving C-3PO.
That may sound like a tall order, but some of the semifinalists have been working on this challenge for more than a year already. The list includes all the teams whose vehicles finished the 2005 Grand Challenge:
- The Stanford Racing Team, whose Stanley SUV won the $2 million.
- Carnegie Mellon University's Tartan Racing, which had two finishers.
- Louisiana's Team Gray competitors, who got their Kat-5 car through the race even though they had to deal with the aftermath of Hurricane Katrina.
- Team Oshkosh, whose monster truck brought up the rear as an official finisher.
Stanford Racing Team |
Stanford's Junior is an autonomous VW Passat.
|
The co-leaders of the Stanford team told me that they're well into their testing schedule for Junior, a computer-controlled VW Passat wagon. They're programming Junior to deal with four-way intersections, parking-lot navigation and traffic jams. "The most challenging maneuver would be merging traffic," said co-leader Mike Montemerlo.
The other co-leader, Sebastian Thrun, told me that this year's race was likely to have more of a "random outcome" than the previous Grand Challenge. As any commuter knows, the best-laid plans to get from point A to point B are easily ruined if you get stuck behind an oafish driver.
"It doesn't just depend on yourself, it depends on the other robots," Thrun said. "If other people act very poorly, then our strategies for dealing with traffic might fail to work."
The win might well go to the vehicle that's best at avoiding traffic tie-ups. "The smarter your robot, the more it will be able to escape," Thrun said.
CMU / Tartan Racing |
Boss is a self-driving Chevy Tahoe SUV, created by
Carnegie Mellon University's Tartan Racing team.
|
Meanwhile, Carnegie Mellon's team members say they have already logged hundreds of test miles with Boss, a self-driving Chevy Tahoe SUV.
"Boss today can handle maneuvers at 30 miles an hour that it performed at 15 miles an hour back in June," Chris Urmson, director of technology for Tartan Racing, said in a news release. "It can park itself and it can yield at intersections, not just stop."
Although Boss and Junior would have to rank among the favorites, the winner could well be Team Gray's Plan B, or Team Oshkosh's TerraMax ... or one of the other 32 teams ... or no one. In any case, Stanford's Thrun is looking forward to one heck of a show.
"I think it's going to be a major event in autonomous driving," he told me.
Here is the full list of 36, linked to team profiles on the DARPA Web site:
