Last year, America's Defence Advanced Research Projects Agency, DARPA, thought it would be a good idea to organize a robot race across the Nevada desert. The idea of the Grand Challenge, as DARPA dubbed it, was for autonomous robot vehicles to steer a 227 km(142 mile) course and claim a $1 m jackpot. This would be a first step towards DARPA's ultimate goal of being able to build unmanned self-driving military vehicles and thus keep American troops out of harm's way on the battlefield.

This year's crop of 23 entrants were offered an even greater incentive—a $2m prize for the winner. That, plus the intervening 18 months, seems to have done the trick. This time, five vehicles finished the 211 km course. The winner, a modified Volkswagen Touareg dubbed Stanley by its makers, a team from Stanford University, did it in a mere six hours and 54 minutes. Stanley was, of course, specially hardened by its designers for the rough terrain of the Nevada desert. The clever bit, however, was the vehicle's brain. This was designed and built by the Stanford Artificial Intelligence Laboratory (SAIL). Stanley's brain consists of six top-of-the-range Pentium chips wired collaboratively together. It is programmed with special software that is able to learn from its mistakes. This software mastered the tricks of collision-avoidance in a series of desert test runs conducted before the race started. Like all brains, Stanley's has a range of sensory inputs to process. A global positioning system (GPS) receiver tells it where on the Earth's surface it is. Television cameras, radar and four laser based distance monitors tell it what its surroundings are like. By comparing its GPS location with its pre-programmed destination (announced only a few hours before the race began), it knew which way it wanted to go. And, by studying its surroundings, it could work out what looked like the safest route that was also in approximately the right direction. Although Stanley carried off the laurels, the other four finishers did respectably. Sandstorm managed a time just ten minutes behind the winner while her sister vehicle Highlander came in ten minutes after that. GrayBot and TerraMax, the other two course-completers, came in at seven hours 30 minutes and 12 hours 51 minutes, respectively. So smart, autonomous vehicles can, indeed, find their way across several hundred kilometres of desert. The question is, what next？ DARPA's answer, of course, will be to go down the military route. But this sort of technology has obvious civilian applications as well, as Sebastian Thrun, the head of both SAIL and the Stanford racing team, is keen to emphasize. Dr. Thrun thinks that it could lead to self-driving road vehicles within 30 years and—more immediately—to greatly improved collision-avoidance systems. Whether the freeways of California will prove as easy to navigate as the gulches of Nevada, though, remains to be seen. The purpose of holding a robot race is to A．adventure through the Nevada desert. B．delevop unpiloted vehicles for military use. C．win a $lm jackpot. D．keep American troops unharmed.