PHOTO: ROBOCUP FEDERATION

T-minus 10 minutes till kickoff, and everyone's set to go—the athletes are finishing their warm-up laps and the officials are huddling in pre-game conferences on the sidelines. Suddenly, the smell of burnt rubber and fried circuits fills the air. This can only mean one thing: A key player has just gone down.

At the 2006 RoboCup (http://www.robocup.org) in Bremen, Germany, these kinds of electronic mishaps were par for the course—along with kamikaze-swift attack passes to rival any Pelé feat. Billed as the tech equivalent of soccer's FIFA World Cup, the event, which took place from 14-18 June, showcased the cream of the artificial-intelligence crop. Thousands of competitor robots from 36 different countries competed in an array of divisions from “small” to “middle-sized” to “four-legged.” The premise driving the RoboCup, now in its tenth year, is that robotic critical-thinking skills are best honed and demonstrated on pool table-sized playing fields.

“In soccer, it's difficult to plan. You can't just say, 'I'll start up the system, and five minutes later, the ball will be in the goal,'” says Hans-Dieter Burkhard, a robotics engineer at Berlin's Humboldt University and vice president of the RoboCup Federation. “You always have to adapt, and that is what humans do very well and what robots are learning to do.”

For members of the RFC Cambridge team, made up of engineering students from Harvard University and the Massachusetts Institute of Technology, adaptation became the name of the game by necessity. With one member of their robot squad out of commission, the remaining three bots were tasked with picking up the slack in their first-round faceoff against the Field Ranger team from Singapore Polytechnic. Initially, the numbers deficit seemed to be too much for RFC Cambridge to overcome. Field Ranger mounted an unopposed blitzkrieg mission as its opening gambit, racking up several goals in less than 5 minutes.

But the tables turned as Field Ranger developed a collective case of performance anxiety. The robots converged around the orange ball in a tense semicircle, the goal within striking distance, but then abandoned their strategy, wandering aimlessly like bored kids waiting for a school bus.

“Was ist das Problem?” one spectator whispered to another.

“Kaputt!” was the reply.

As similar confused moments on other courts illustrated, the biggest challenge for many teams has been designing robots that can respond when unexpected situations crop up. In order for a computer-based brain to be truly adaptable, “its internal world models have to be very complete,” says Anna Foerst, an artificial intelligence expert at St. Bonaventure University, in New York. “If not, it might encounter a situation that’s completely new, and it might not have a parallel for that in its memory.”

The Attempto team from Germany's University of Tubingen hit just such a snag in one of its early matches when the sun streamed through the windows of the arena. “When sun comes in from the outside, the colors on the field change,” says team member Hannes Becker. “That can interfere with the robots' visual systems.” If the bots can no longer detect the wavelengths of shades they have been precisely trained to recognize—the green of the turf and the orange of the ball, for example—coordinated drives toward the goal lose their momentum.