Will cars ever be capable of driving themselves? Someday. But the computer software packages designed to control steering, braking, and throttle are in the midst of a trial-and-error learning stage all too reminiscent of a teenager's first experience behind the wheel. Only after some unnerving instruction—and perhaps a dented bumper or two—are they good enough to go solo.

On 9 October, computer algorithms showed that cars might just be ready to take the wheel without human chaperones. That was the day that four autonomous vehicles completed a 211-kilometer racecourse stretching through Nevada's Mojave Desert in less than 10 hours, as required by the rules. The autos avoided boulders and other obstacles, traversed bridges, and maneuvered through hairpin turns on mountain switchbacks as they vied for the US $2 million winner's purse.

The race, organized by the U.S. Department of Defense's R&D arm—the Defense Advanced Research Projects Agency (DARPA)—was won by a Volkswagen Touareg SUV developed by a team from Stanford University, in California [see photo, "Winner"]. The Stanford car, dubbed Stanley, finished in 6 hours, 53 minutes—11 minutes ahead of the second-place finisher, one of two Hummers in the race that were rigged up by the Red Team from Carnegie Mellon University, in Pittsburgh.

The outcome of the race, known as the Grand Challenge, showed just how far autonomous vehicles have come in a year. No one claimed the $1 million prize offered by DARPA in 2004. That year's winner—if you could call it that—hadn't gone 12 km before it skirted the edge of a cliff so closely that it got stuck. Its wheels continued to spin until one of the tires caught fire and the vehicle had to be deactivated remotely. Only two of the race's other 14 entrants passed the 2-km mark [see "Sand Trap," IEEE Spectrum, June 2004].

Explaining the vast improvement over last year's pitiful showing, a Stanford engineering school spokesman said: "The robotics community has learned a great deal about how to make cars drive themselves. In fact, artificial intelligence was employed in Stanley's programming that allowed it to continually learn during the testing conducted in the months leading up to the race."

DARPA issued the challenge as a way to help the Defense Department meet a 2015 deadline for making 30 percent of the U.S. military's land vehicles autonomous. Unmanned vehicles could prove valuable in combat zones such as Iraq, where hundreds of soldiers and civilians have been killed or maimed by explosive devices planted on roadbeds.