Slashdot Mirror
Robocup 2002 World Robot Soccer Championships
dipfan writes "While the rest of the world is watching the soccer world championships in Korea and Japan this summer, at the same time the Robocup 2002 competition for soccer-playing robots is to take place there - the sixth time the tournament has been held, with 35 countries competing and this time including a "humanoid league" competition. The purpose is to foster research in robotics, with the aim of building a team of robots that can play and win against the best human teams by 2050. One of the pre-tournament favourites this year is Iran, who did well in 2000 but not in 2001. The Swedish team includes a star player named Priscilla, described as "looking like a sister of the Terminator". One of the Swedish designers comments: 'you don't want to give too much freedom to the robots as they will go crazy.' Much like flesh-and-blood highly-paid sports stars really."
(Note: I'm sitting in a RoboCup lab right now, so IMNSHO:)
:) There are a lot of strong teams in the league though so we'll have to see...
:) It sure did that for me, even though I've been programming them for several years.
I don't think the favorite is going to be Iran this year, but more likely the Phillips professional team, which won the German Open this year. That said, I wish people would realize there are 4 leagues, not just the middle size league, with different robots and different favorites in each. In the Sony Legged league, UNSW has dominated, though we came in second
In the small size, I'd say the favorites are last year's winner LuckyStar II from Singapore, and Big Red from Cornell University. FU-Fighters is also a pretty strong team. Our team (CMU) hopes to do a lot better this year in the small size league. We won in '97 and '98, but haven't done too well since then.
I don't know to much about the simulation league so I won't bother to comment. Finally, a personal plug: See a video from the vision system of a Sony legged robot here. It'll give you more respect for how hard a problem this is
That would be the FU-Fighters 1999 team. It was a spinning 5mm thick aluminum plate, which is pretty large considering the robot itself was about 10cm x 12cm. They could kick the ball at up to 6m/s. They turned it down later in the competetion, but it was not banned (more to keep from getting penalties than other team's anger).
The design lives on still in a toned down version on the current FU-Fighters team and others have copied it too. It's affectionately known as a "spinning blade of death" kicker.
Because it's an enormously difficult task to do (as a robot) what you and I take for granted.
Motion (walking, for example) involves all sorts of attitude shifts to maintain the centre of gravity in the correct place - it's not just putting one foot in front of the other. If you look at how athletes use their bodies to kill momentum and do quick turns, I think you'll agree that the slow-plodding robots are a long way off yet...
Machine vision is still in its' infancy. What you have is unsurpassed. What a 95% blind man has is unsurpassed by machines. Vision implies cognition, (without it, it's just a TV screen!) and this is Hard (capital H). Being able to recover depth, handle occlusion and parallax, make inferences from absolute and relative properties (size, circularity, colour, etc.), and generally 'see' is simply not possible yet in the general case.
Temporal tracking is needed to tie each of those frames together, and make further inferences and aid cognition. The ball moves, after all... kicking it is actually solving a pretty complex tensor of vectors that are constantly changing...
Prediction is necessary to make best use of the current tactical position, with allowances for what might happen next. If your defence has just got the ball, the forward might want to make a run towards the opponents goal... This sort of thing is 'common sense' to people, but really hard to get a computer to come up with. It's easy to hard-code some rules into the machine, but the real goal (pun intended!) is to get the machine to devise its' tactics from instant to instant.
In short, although it's "just a game", it embodies a large cross-section of really hard problems in robotics. Frankly, my money's on the humans!
Simon
Physicists get Hadrons!
I'm on the University of Virginia team, and we're one of two teams to qualify representing the US for the simulation league (which doesn't use real robots, and is thus a lot more fast paced). The other team being AT&T Research Labs.
We're actually in the process of looking for sponsors so that we can get to go to Japan for the competition.....
Big mistake.
The book is free. It is entirely available on the Internet, and no other place.
The book merely shows that I think deeply about the subject of adversarial behavior. That's the reason I mentioned it.
I'm on the University of Virginia team, and we're in the simulation league. We don't have physical robots. The result is we don't have to spend the effort on mechanical concerns and image recognition, and we get to devote more effort to the AI aspect of the competition. In the simulation league, the robots DO get tired, they do miss shots, and they have imperfect information. The idea is to remove all physical considerations from play and have them win on intellect alone. The robots being used in competition are no where near the level required to beat a world cup team, but I imagine that once they get close, these "humanizing" factors will be added to keep the match fair. At the current rate of development, I expect this to be well before 2050.
WARNING: there is a trojan on your