RoboCup 2013: Team Water Is Middle Size League World Champion

An anonymous reader writes with a bit of sports commentary on the just-concluded RoboCup 2013 soccer matches: "Previously achieved results are no guarantee for the future, as was demonstrated once again in the final match of the Middle Size League. Team Tech United Eindhoven had reached the final unbeaten and without a single goal against them, but the Chinese team Water turned out to be the stronger party in the final." It's hard to stop watching video of soccer-playing robots.

Comments

More accurate video

[Ken_g6]

I couldn't find video of this year's match, but here's video of last year's. The robots in question are not humanoids, and not like those shown in the video in TFS.

Re:More accurate video

[Anonymous Coward]

Right, there are six leagues that play actual soccer or something more or less resembling it at Robocup

Small Size League are tiny cylindrical wheeled robots with no onboard vision. They all "see" through a camera mounted above the pitch that looks down. They zip around really fast, shoot their tiny ball even faster. They are interesting for developing tactics, but they're mostly a dead-end from the mechanical POV at this point. You won't be bored watching this (at least, so long as they aren't taking 10 minute breaks to fix mechanical problems) but it's pretty hard to follow because they move so fast, I don't know how the referee checks if something was a goal.

Medium Size League are bigger wheeled robots, with onboard vision playing something that looks a lot like human 5-a-side soccer. They go pretty fast, they roll the ball in front of them with a special mechanism, rather than "dribbling" like a human player, and they can "kick" it using the same mechanism either along the ground or up in an arc hoping to maybe get over defenders and into a busy goal. That's the league Water was in. It's as fun to watch as a kick-around soccer game IMO.

Next there are three sizes of "humanoid league", kid-size, teen-size and adult-size. They roughly correspond to the minimum permitted sizes in each league, making a taller robot is harder, because bipedal forms are not very stable, so at "adult" size the state of the art is one robot per side basically slowly walking towards a ball, and kicking it with a significant chance of falling over when doing so. At kid-size they're not quite running, but they're maybe jogging or something, and they play something closer to actual soccer in teams. Teen-size is somewhere in the middle.

Finally standard platform league uses an off-the-shelf (almost) commercially available robot. So in this league it's purely about programming, you can't have better motors, better sensors, cleverer mechanical design, you have to work with the generic "Nao" soccer robot. This league is a similar scale to "kid size" but it's more competitive now because not needing to design and build the robots frees up more time for AI work. These robots play 5-a-side, albeit on a smaller pitch than the MSL. You can see a really human-like formation, with a defender and goalie holding back even as 2-3 attackers go up front to press the advantage. It's less exciting than MSL but the finals at least (where the best teams play each other) are worth watching IMO.

To get any idea of how we're doing on the problem of playing against humans, there should be footage of the MSL winners (Water) playing the Robocup trustees (so basically a bunch of middle aged but still physically in OK shape academics). Ten years ago this was a joke, a group of kids could beat the robots soundly and grew bored doing so. Today a healthy adult has to put some work in to keep up and get one past the robots. The humans have better reflexes but lack accuracy and make too many dubious snap decisions. Ten years from now? I doubt the academics will win against the wheeled MSL, if it still exists.

Re:More accurate video

[harishamdani]

this is their official site though, maybe this helps : http://www.robocup-2013.org/

Pro leagues are different

[Anonymous Coward]

The USA team, with two ED-209 units acting as midfielders/enforcers, is intimidating the competition.

losing a soccer final is a dutch tradition

[Anonymous Coward]

The Dutch traditionally lose the final to a world championship after having very strong games leading up to it (1974, 1978, 2010). So the Eindhoven team performed exactly as expected. It's very impressive.

Water? Cheaters!

[Anonymous Coward]

Wait. If I understand the summary correctly, they're implying that they beat the other team by dumping water on their circuits? :P

I for one welcome our new water-resistant soccer-playing robotic overlords!

Progress can be subtle

[Anonymous Coward]

If you watch videos from different years, you might get the impression that the robots aren't getting better.

Actually though because this is about driving research, rather than putting on a fun spectacle first and foremost, the status quo is caused by them making the rules harder, so the robots (or rather, their engineers) are constantly catching up to more difficult rules. The pitch has been slowly growing, the image recognition task is becoming more realistic (coloured position markers went away, now the goals are the same colour at each end) the humanoid leagues have been shrinking the foot size (which makes walking stability far harder, a human has _tiny_ feet yet it can somehow run over uneven terrain, the robots must some day match that) and they reduced vision to a human-like 180. The teams grow bigger (which means more co-ordination and greater chance of running into another robot) and the trickier rules of soccer like off-side, corner kicks, and the like are slowly being introduced too.

Highlights of the final

[cablepokerface]

... Are here: http://www.omroepbrabant.nl/?video/82273612/Hoogtepunten+uit+de+finale+Robocup+2013+Tech+United+-+Water.aspx&action=media It's a dutch site. But just press play on the video player.

MOD PARENT UP

[Katatsumuri]

Good description of the different RoboCup leagues; too bad parent posted as AC.

The hardware is getting better

[Animats]

The Nao is a quite capable little robot. It costs $16,000, though. (There are promotional discounts, developer discounts, academic discounts, etc. But that's the list price.) The lowest priced good humanoid (the Bioloid) is around $1200, so these things are approaching affordability in half-meter size. The low-end robots use improved R/C servos (ones that talk on a bus and provide useful feedback info). Nao has custom mechanics and even a 3-fingered hand.

The locomotion control of the little guys is still rather disappointing. They're mostly still at the "walking on big feet" level. The actuators and inertial sensors are good enough for dynamic running, and there's enough processing power, but the control theory and software aren't there yet.

It used to be that if you went to a major academic robotics lab, none of the robots were running. Now, there will usually be something going.

Oh boy!

[slick7]

I can't wait to see what they do on the battlefield!

Re:Oh boy!

[Richard Kirk]

When I looked at the video, YouTube also offered a trailer for 'Pacific Rim'. Oh, how we laughed.

Re:Oh boy!

[Ksevio]

Presumably they'd play soccer on whatever field you put them on as long as it's flat enough.

So a battle field would probably be too rough and they'd fall over.

Re:Oh boy!

[slick7]

Presumably they'd play soccer on whatever field you put them on as long as it's flat enough. So a battle field would probably be too rough and they'd fall over.

Until they learn to jump.