Published online 6 October 2005 | Nature | doi:10.1038/news051003-9

News

Robots rev up for Grand Challenge

They're back, and this time they actually work.

Stanley, the Volkswagen Touareg entered by California's Stanford University team, came second in the trials.Stanley, the Volkswagen Touareg entered by California's Stanford University team, came second in the trials.© Stanford University

Robots, start your engines. A US$2-million prize awaits the first autonomous vehicle to complete a high-speed cross-country trek this weekend, over some 240 kilometres of rough terrain.

The Grand Challenge race around Primm, Nevada, is organized by the US government's Defense Advanced Research Projects Agency (DARPA). It aims to find robot vehicles that can drive around war zones reliably without risking the life of a driver.

The first Grand Challenge in March 2004 ended in disappointment, with none of the competitors completing the course. In fact, only four vehicles managed to make it more than 8 kilometres before crashing and, in some cases literally, burning.

H1ghlander (left) is the bot to watch. The Red Team's earlier model, the Humvee Sandstorm (right), is running third.H1ghlander (left) is the bot to watch. The Red Team's earlier model, the Humvee Sandstorm (right), is running third.© Carnegie Mellon University

Most of the teams have spent the last 18 months improving their vehicles' artificial intelligence. The result has been impressive. "The entire field in the 2005 Grand Challenge has advanced significantly beyond what we saw on 2004," says DARPA director Tony Tether, who doubled the prize money for this year's competition to attract more entrants.

First hurdle

In last year's trials, some robots failed even to make it off the starting line, but 2005 saw more than half the competitors tear through the 4-kilometre test course in as little as ten minutes.

Big and small: TerraMax (right) is the biggest semifinalist, while Ghostrider (left) didn't make the cut.Big and small: TerraMax (right) is the biggest semifinalist, while Ghostrider (left) didn't make the cut.© DARPA

After the first day of trials, the Grand Challenge programme manager Ron Kurjanowicz said: "What some of these teams were able to accomplish today would have been unthinkable a year ago."

The week-long trials ended on 5 October, and DARPA swiftly announced the 23 finalists who will line up on the starting grid for the race proper on 8 October. The field contains some familiar faces from last year, and many new ones.

Smarter, faster, stronger

DARPA officials took a lot of care in selecting 43 semifinalists from 195 applicants for trials at the California Speedway. They demanded extensive technical specifications and made inspection trips to the teams' garages.

The robots use an array of cutting-edge technology to navigate, including global positioning systems, radar, laser range-finding and stereo-camera vision. All carry impressive computing power to filter data so that the vehicles can find their way while avoiding obstacles, cliff edges and traps. In order to complete the trek within the required ten hours, the vehicles will have to average about 30 kilometres per hour.

Spider is another bot raring to go for the race this weekend.Spider is another bot raring to go for the race this weekend.© Cornell University

The hot bot to watch is H1ghlander, a converted H1 Hummer from the Red Team, based at Carnegie Mellon University in Pittsburgh, Pennsylvania. Managed by robotics expert William 'Red' Whittaker, the team has seen its earlier model, the Humvee Sandstorm, win the third slot on the grid.

Last year, Sandstorm beat off all competition by making it a full 12 kilometres through the desert before getting stuck on a boulder and catching fire.

Stanley, the Volkswagen Touareg entered by California's Stanford University team, will take second place on the grid. It was the only car to complete four runs through the training course without hitting a single obstacle, which its handlers attribute to the six Pentium-toting computers mounted in the trunk.

“What some of these teams were able to accomplish today would have been unthinkable a year ago.”

Ron Kurjanowicz
Grand Challenge program manager

"Since last fall, dozens of Stanford faculty and students have been working to write, test and refine the code that makes Stanley able to find the road and make sound decisions," explains Mike Montemerlo, part of the Stanford team.

The largest contender is TerraMax, a modified military truck from Ohio State University in Columbus that performed relatively well last year. Although slower than the rest of the field, TerraMax has an edge because it can cope with steep gradients and drive through rivers.

Ghost of a chance

Not all the contenders were so lucky. Ghostrider, a riderless motorbike from University of California, Berkeley, made a second failed attempt at the trials this year.

ADVERTISEMENT

Back in 2004 it failed to stay upright for more than a few seconds, but the team has clearly made some progress by adding a secret weapon. After an unfortunate collision with a metal barrier earlier this week, the plucky two-wheeler simply popped out a pair of struts that bounced the vehicle upright again, allowing it to get through more than half the trial course. But no more.

The full race will prove much more challenging than the trials, taking in bumpy desert roads, dry lake beds and narrow mountain passes. The teams will only learn the route two hours before the race begins, so that they have no chance to preprogramme their robots.

DARPA promises that if a winner is not found on 8 October, it will double the purse once again, offering a sweet $4-million prize in next year's race. 

Grand Challenge program manager