Competitions speed up advances in robotics

Wired on Friday: The world of robotics is one of those pursuits that is peppered with cheerfully optimistic predictions that are then quietly buried for decades after their supposed due date, writes Danny O'Brien

Robot menservants by 1976? Cars that drive themselves by 1984? Let us not speak of such things.

But, in a highly unusual turn of events, the past 12 months have seen the now cautious roboticists somewhat surprised by success.

Take this year's Darpa grand challenge. This is an annual competition for designers of autonomous vehicles - cars that drive by themselves - which is held by the US government's Defence Advanced Research Projects Agency.

The challenge seems relatively simple if you're a human: get a car across a 150-mile track in the Mojave desert in less than 10 hours, without crashing into obstacles or wandering off the prescribed route.

When there's no-one in the driver's seat, however, things can get hairy very quickly. Until now, the barrage of blurry, ambiguous information that floods into a car driver's consciousness, and the split second decisions that have to be made even at low speeds, put the task far away from modern, clunking, self-contained robots.

The robot cars are designed by experts from commercial companies, academia and enthusiasts, but no-one is allowed to touch them once they start their desert trek.

And robots, it is held, aren't quite mature enough to go out alone. Indeed, the first time anyone even vaguely considered this to be possible with modern technology was last year.

The 2004 Darpa grand challenge happily lived down to everyone's low expectations. Out of a field of 15 (whittled from a 106 teams), only three made it past the first mile. Those three all died within eight miles. One got stuck in the wrong gear driving up a hill; one fell into a ditch, and the other set its own tyres on fire.

The general opinion, from attendees I spoke to, was that the whole thing was sometimes embarrassing, often hilarious, but never seriously expected to be completed.

This year was different. Five robot cars made it down the entire route. The winner, a roboticised VW Touaref kitted out by a group of Stanford computer scientists, did it in six hours and 53 minutes, at an average of about 25 miles per hour.

So can we expect to see those self-driving cars on our streets? Unlikely. Navigating a desert track is a very different problem to driving through busy streets full of soft, fleshy humans.

But it is closer - deliberately closer - to another set of problems. Darpa's unashamed intent with the grand challenge is to drive innovation in the practical field of autonomous military vehicles. Scouting, retrieving and possibly even leading attacks in hostile terrain is what Darpa has in mind with this technology.

The morality of that intent is questionable, especially if you're sympathetic to the fleshy humans that lie in the general direction of the US military's sights. But this research is certainly driving forward the frontiers of this area. And Darpa is, commendably, taking the same innovative approach to encouraging solutions to these problems as it did back when it seeded the investment to create Arpanet, which would later become the internet.

The Darpa challenge was an entirely public event, with a tempting $2 million (€1.65 million) of taxpayers' money as a prize. Stanford took that money home, with only a $500,000 investment (much of which was paid for by sponsors).

But for that $2 million, Darpa got to see more than 20 different approaches to the problem.

And it's not as if such autonomous research doesn't pay off in other areas, and in other countries. Up in the air, unmanned vehicles are being used to glide through hurricanes and, in Israel, they are even being used to catch speeding violators.

The UK has recently given the green light for considering civil autonomous aircraft in its airspace.

None of these crafts are entirely computer-controlled, but they're certainly more than large radio-controlled models.

Back in the army, one pilot last month was able to control a sortie of four aircraft simultaneously, as they weaved over the same Nevada landscape as the Darpa challengers.

After years of slow progress, we're suddenly learning to create crafts that can be independent and survive on their own in harsh conditions.

But perhaps the lesson here is more to do with the means than the end. The same week as the Darpa challenge across the hot dry lands of America's southwest, 20,000 spectators watched a simple rocket fly across the sky, operated by an air force colonel. The scene was the countdown to the X-Prize cup expo, an air show for spaceships.

The event isn't a competition, but is sandwiched between two privately funded competitions, similar to the Darpa challenge. The first was the X-Prize, a $10 million prize awarded last year for the first private sub-orbital rocket trip.

The next will be the X-Prize cup, a grand prix for sub-orbital ships, with rewards for fastest turnaround between two launches, maximum number of passengers, maximum altitude and fastest flight time.

As the Darpa grand challenge has driven forward robotics, the X-Prize series has driven forward private space flight.

These public competitions - privately or government-funded - are doing a fine job of pushing autonomous craft on earth and manned craft in space that last mile, just as prize competitions did in the age of air-spurred figures like Charles Lindbergh to fly across the Atlantic.

We're only beginning to learn how to drive robots across land and through air, and perhaps even to use them to ferry people into space. But we do at least have some real experience at knowing what drives human achievement.