By LOREN NEWMAN

STAFF WRITER

Last October, Stanley, a Volkswagen Touareg R5 modified by the Stanford Racing team to drive autonomously, became the first robot ever to complete the 131-mile Defense Advanced Research Projects Agency (DARPA) Grand Challenge. DARPA, the central research and development agency for the U.S. Department of Defense, began hosting the robotic race in 2004 in an effort to push private researchers to develop robotic technologies.

Now that the Grand Desert Challenge has been conquered, DARPA announced a new Urban Challenge on May 1. The Stanford Racing Team said it plans to defend its title.

“This race is about urban driving. The types of problems are merging into traffic, behavior at a four way stop and interacting with pedestrians and other moving vehicles,” said David Stavens, a PhD student in computer science who was a key figure on last year’s team. “It’s very different from the first grand challenge, which was held entirely in the desert.”

Urban environments provide a whole host of new challenges for developing a robot that can successfully navigate an area.

“The biggest technological problem is being able to see moving obstacles,” Stavens said. “Previous perception technologies only allowed us to see stationary things. You have to be able to make a prediction about where the obstacle is going to go.”

The team plans on using one of three Volkswagen Passat sedans donated to the project by Volkswagen. Stanley — the robot who won last year’s challenge — is scheduled to be retired after a grueling year of desert testing.

“It will be outfitted very similarly to Stanley, but with some new twists,” Stavens said. “You’ll see actuated [moving] lasers on the car this time. Also, to drive at high speeds, you need to see farther, so there will be higher-powered lasers on the car. We’re partnering with APPLANIX. They build completely integrated navigation systems for cars — like GPS but a lot better and at a higher resolution.”

Though early press releases from DARPA indicate that a city of some kind will be used for the race, team members were skeptical of the plausibility of running a hundred driverless cars through a real city.

“We don’t know for sure what city will be used for the race,” Stavens said. “I would be surprised to see it in a real city, just because of the liability issues.”

Computer Science and Electrical Engineering Prof. Sebastian Thrun, director of the 150-person Stanford A.I. Lab, will once again head the group.

A press release from DARPA describes the Urban Challenge as a “simulated military supply mission in a mock urban area” which will be 60 miles long. Teams will be expected to complete the course in less than 6 hours. The race, like last year, features a top prize of $2 million with a $500,000 second place award and $250,000 for third.

The challenge has practical applications as well. According to Stavens, 43,000 people die per year on America’s roadways, due in large part to driver error.

“We hope that a smarter vehicle, one that is more aware of its surroundings, can warn the driver or take some kind of action to save lives,” he said. “DARPA obviously has military applications for this technology, and they will pursue that. But we’re participating in the challenge for the humanitarian component of it.”

There are two tracks for competitors to take in next year’s grand challenge. One is very similar to last year’s procedure, which involves primarily private funding; the other track provides an opportunity for teams for apply for up to $1 million in research grants from the government. The Stanford team intends to apply for the research grants to fund their car in the upcoming challenge.

“In the past, a lot of schools were hesitant to jump in full force and participate because it’s a big risk, you have to raise a lot of money and there’s no money up front,” said Stavens. “This year, with the grants, more strong robot-builders will compete who may not have in the past.”

Last year 195 teams competed in the desert. Stavens predicts a jump in entrants due to the availability of research grants. The Stanford team is currently gearing up for the long trial.

“We’re just getting started. DARPA has announced the challenge; we’re trying to read through the rules to get a sense of what the challenge is and what we’re going to have to build,” Stavens said.

The team’s Volkswagen Passats will arrive in June or July, two for testing only, and one for the race. By July or August the team will be ready to start working on the computer algorithms in the car. Stavens said the team hopes to be doing autonomous city driving by the end of 2006, using the next year to clean up loose ends.

There will be a national qualifying event in late October 2007 and the final race is scheduled for Nov. 3, 2007.

“This is much harder than the first Grand Challenge,” Stavens said. “The city is much harder scientifically. It’s a little more nerve-wracking, but we’re scientists and we are excited about attacking a really hard, state-of-the-art problem. We will be strong. It’s hard to say if we’re the favorite, but we’re optimistic.”