May 21, 2002
In the segment "World Cup for Robots," Alan watches real, nuts-and-bolts robots play a live soccer game. The physical matches of the RoboCup league are only a part of the competition, however. Equally important is the "virtual" simulated match, where artificial intelligence can be developed in a pure computing environment, free from any mechanical constraints. Here you can have a look at a couple of recent simulated championship matches.
- - - - - - - - - - - -

The Cutting Edge of A.I.

Bearing a passing resemblance to a video game from the early 1980's, the simulated RoboCup matches don't look very impressive - one could be forgiven for assuming that the underlying programming is similarly simplistic. Despite this austere appearance, however, the simulated RoboCup platform is actually a test bed for the cutting edge of artificial intelligence and robotics research. "The goal of the simulator and of the real robots is to make it as realistic and challenging as possible, so that the lessons learned are as general as possible," says Peter Stone, Computer Sciences Professor at the University of Texas at Austin and former "coach" of Carnegie Mellon University's simulated team, CMUnited. In 1999 he and team members Patrick Riley and Prof. Manuela Veloso led CMUnited to victory in the simulated RoboCup championship.

Simulated RoboCup is very different from the graphically lavish, almost photorealistic video games playing on televisions in living rooms across the planet. In a familiar console-style video game of a soccer match, the computerized "players" are part of a single larger program that possesses complete information on the positions and states of all the objects within the game. In contrast, RoboCup soccer models a more realistic situation where the individual players have only a limited view of their virtual "playing field," and must make decisions independently, with imperfect knowledge of what's going on around them. The members of the competing teams in a RoboCup match are individual programs that must be equipped with both a general strategy, so they can work with other teammates effectively, and enough individual smarts to make decisions and take initiative based on the situation, just like in real life.

	The simulation league allows programmers to test the latest in artificial intelligence. See some examples.

Stone calls his approach to programming a "locker-room agreement" - a way for the programmed teammates to share plans and signals before the game so they can coordinate their actions with only minimal communication during the actual match. The players in the simulated matches represent idealized robots, but the algorithms that are developed here can be applied to the real thing, often with winning results. "Since the locker-room agreement deals mainly with the high-level concepts of playing soccer, and not the low-level mechanics or skills issues, I was able to transfer it to a real robot team with about a half-hour's worth of coding," Stone says.
- - - - - - - - - - - -
2 pages: | 1 | 2 |