A Cary-based Lotus sports car dubbed Lone Wolf is one of 35 robot vehicles that will meet today in California for the semifinals of the $3.5 million DARPA Urban Challenge, a Defense Department contest in driverless driving.

The Defense Advanced Research Products Agency is holding six days of qualifying trials at the former George Air Force Base in Victorville, Calif. The best 20 vehicles will stay there for the final 60-mile competition Nov. 3.

Packed with computers and video, radar and laser sensors, the autonomous cars and trucks will navigate urban streets, dodge live traffic and squeeze into parking spaces.

When obstacles block their paths, the robot vehicles will have to check their GPS maps to plot detours. Along the way, they must obey California traffic laws.

In the Nov. 3 showdown, each car will have six hours to finish four simulated military supply missions totaling 60 miles. The top three winners will receive prizes of $2 million, $1 million and $500,000.

Some competitors have invested much more than that, with corporate-university partnerships dedicating full-time engineers to the effort. Carnegie Mellon University's team, backed by General Motors, built a backup clone for its 2007 Chevy Tahoe. Both robot SUVs learned their street smarts at GM's Desert Proving Ground in Arizona.

Lone Wolf is the creation of Insight Racing, a team of 50 volunteer engineers and N.C. State University students.

Grayson Randall of Cary, 52, a senior software engineer at IBM, is the group's leader and the car's mastermind.

Randall figures it would cost about $300,000 to replicate the little blue sports car and all the hardware provided by Insight Racing's corporate backers.

"That doesn't include any software," Randall said. "We've got many man-years of [unpaid] software development, and that's really the key to our work."

Under a congressional mandate to develop autonomous ground vehicles for future war zones, DARPA announced its robot competition in 2003.

Fifteen vehicles started the first race in 2004 on a 142-mile desert course, and none finished. In 2005, Stanford University's car won $2 million as the first of four finishers of a 132-mile desert race.

Contestants and DARPA officials agree that the urban traffic competition will be more difficult than the desert races, but the robot cars will be smarter than ever.

"If somebody came to me five years ago and said there will be cars driving themselves, I would have said it was science fiction," said Simon Cobb of Lotus Engineering, an American subsidiary of the British car company, which provided the car for Insight Racing.

"But there are going to be 35 of them in California, driving themselves, and it isn't science fiction."

Software will determine how Lone Wolf's computers read the car's surroundings and make quick decisions -- moving at speeds up to 30 mph -- about how to negotiate traffic and reach the finish line.

"Whoever can handle the most situations and recover from errors and problems is going to do the best. I think our software is strong, and we have a great opportunity to show that we can handle most of those situations," Randall said.

"Hopefully, heh heh, all of them."

Get $150+ in coupons in every Sunday N&O. Click here for convenient home delivery.