Less than 48 hours after one of his team's robotic race vehicles, H1ghlander, rolled over during a practice run last Monday in Nevada, William "Red" Whittaker remained upbeat about the Oct. 8 Grand Challenge off-road robot race.

	Red Team photo One of two self-driving Hummers being prepared by Carnegie Mellon University's Red Team, during testing at the Nevada Automotive Test Center near Carson City, Nev. Click photo for larger image.

"On its worst day," he explained as he chomped on a lunch of Chinese food in his Carnegie Mellon University office, "H1ghlander's performance is very strong relative to the Grand Challenge field."

As he ate, members of his Red Team, 2,100 miles away at the Nevada Automotive Test Center, labored to fix the H1 Hummer, which has been modified so it can drive itself. Within hours, the vehicle was back in limited operation.

But if H1ghlander's chances remain good, the chances are excellent that some vehicle will emerge as the winner of this year's Grand Challenge and claim the $2 million prize offered by the Defense Advanced Research Projects Agency, Whittaker said.

Up to 43 vehicles will gather Wednesday at the California Speedway outside of Los Angeles to vie for 20 starting slots in the race, which will begin and end outside of Primm, Nev. To win the $2 million prize, a robotic vehicle must complete a course of up to 175 miles across the Mojave Desert within 10 hours without any human intervention.

"Two years ago," when DARPA first issued the challenge, "a fundamental question was whether the Grand Challenge was beyond our reach," Whittaker said.

The doubts were justified during the first run of the race last March. No vehicle went farther or faster than the Red Team's modified military Humvee, Sandstorm. And Sandstorm covered only a little more than seven miles of the 142-mile course before stranding itself on a berm.

But Whittaker is confident that some vehicle should emerge victorious next month; both of the Red Team's entries, H1ghlander and Sandstorm, have traveled 175 miles under race conditions during testing this summer in Nevada.

	Red Team photo H1ghlander, one of two self-driving Hummers being prepared by CMU's Red Team, avoids a tank trap during testing in Nevada. Vehicles will have to avoid such obstacles during the Oct. 8 Grand Challenge race sponsored by the Defense Advanced Research Projects Agency. H1ghlander rolled over and was damaged during testing a week ago. Click photo for larger image.

"That's a sea change," Whittaker said of the improved performance. Now, it's just a matter of which team's machines will be durable enough, smart enough and, frankly, lucky enough to win the prize. "There's nothing shabby about any team that's in this thing."

"The prize will be claimed this year, that's my prediction," said Sebastian Thrun, head of Stanford University's team. The Stanford Racing Team's entry, a roboticized Volkswagen Touareg called Stanley, has been routinely making 100-mile drives all summer.

The Stanford team repeatedly ran the first Grand Challenge route between Barstow, Calif., and Primm until July 29, when DARPA officials ordered all the teams out of that area. So many teams were practicing there that the federal Bureau of Land Management complained about damage to the delicate desert.

Even before they were ordered out, Thrun said his team already had been contemplating a move elsewhere because the vehicle's learning software had gleaned about as much information as it could from running those trails. Since then, the team has been running Stanley through the Sonoran Desert outside of Phoenix.

"At this point, we're most concerned about the vehicle itself," said Thrun, a Carnegie Mellon computer scientist before he joined Stanford two years ago. So the team, which includes former CMU grad student Michael Montemerlo, scaled back its driving last week to reduce the risk of a mishap such as the one that befell H1ghlander.

Whittaker insists that H1ghlander and Sandstorm were running fast, but not nearly as fast as they could. Peak performance actually was reached about three weeks ago; since then, the team stopped pushing to go faster and has concentrated on getting it to run fast enough to win.

Driving performance does degrade as the distance increases, however, Whittaker said, noting H1ghlander rolled after 141 miles of desert driving. "These machines take a pounding," he said, so nuts, bolts and connectors come loose and dust infiltrates every nook and cranny during the long runs.

Unlike last year, when competitors were told to plan for a race course as long as 250 miles -- necessitating an average pace of 25 mph to complete the course within the 10-hour time limit -- DARPA has reduced the maximum race length to 175 miles, making it possible to complete with an average pace of just 17.5 mph.

Even TerraMax, a lumbering, 15-ton military truck that is the Oshkosh Truck Corp.'s entry, has been able to average 19 mph over a 20-mile test, the company said. Last year, TerraMax crept along until it was stopped dead in its tracks by a bush that looked insurmountable to its sensors.

DARPA officials, worried that some competitors might simply navigate using global positioning coordinates, have promised to throw some obstacles, such as tank traps, in the path of the robots to ensure that they are able to respond to changing conditions.

The Red Team already has shed some of its sensors; a stereo vision system was scrapped when it didn't work as well as needed, as was a single-camera navigation system.

Both Sandstorm and H1ghlander will rely primarily on a laser radar, mounted on a sophisticated gimbal arm that isolates it from ground shocks and allows the vehicle to look in the direction it is turning. They also use laser range finders, though even some of those are being jettisoned.

Sandstorm will use radar for looking long distances and through dust, though Whittaker said he may not bother to remount the radar on H1ghlander.

It's not that these systems don't work -- "We can get all this stuff to work" -- but whether they are reliable. A sensor plagued by false-positives -- indicating something is in the vehicle's path that isn't really there -- can cause more confusion than it's worth, he said.

DARPA is sponsoring the event in hopes that it spawns innovation in self-driving vehicles, which promise to play a growing and prominent role for the military. But Whittaker sees its impact as more profound than that.

"It will shift the world's view of what's viable," he said. It's not just military vehicles that will take advantage of these technologies, but a wide variety of consumer and industrial products. Robotic vacuums already are on the market, but self-driving lawn mowers and tractors are on the horizon. Honda and Opel have announced they will begin offering cruise control that actively steers a car to stay in its lane as early as 2006.

"There's no putting this genie back in its bottle," Whittaker said. And the Grand Challenge has succeeded in attracting at least 1,000 "rock solid" researchers to the field.

"We are building the next generation of 'Top Guns,' " he said.