Update- probably should have seen this coming: the video had ‘Eye of the Tiger’ as background music and YouTube pulled it. The new version above is on mute, which makes it 10x less awesome. If iCub’s trainers are reading this, you can add the track directly through YouTube legally, but until then, everyone else will have to click here and just let the second video run in the background.

No, I have absolutely no idea why iCub is being tasked with a work out, but I do know why his trainer is berating him: those are sissy push-ups! C’mon, get those knees off the ground and keep your robobutt down! Otherwise, how are you gonna work off all that, uh…

…Flab?

You know, it’s probably worth mentioning that exercise is actually counterproductive for robots, in that it’s likely to make their servo motors weaker over time. But that’s not the point! iCub hasn’t bulked up in years, so it’s high time this robotic baby gets to work.

[ iCub ]

Well, this is pretty awesome:

You’re actually watching the extent of iCub‘s learning process: it took the robot all of 8 trials to figure out how to hit the center of the bullseye. iCub is using a learning algorithm called ARCHER (Augmented Reward Chained Regression), which is optimized for tasks that have an easily definable goal and measurable progression towards that goal. Basically, hitting the center of the target equates to a maximum reward, and the algorithm builds off of past experience to estimate how to alter iCub’s hand positions to improve the aim of the arrow. In this case, the distance between iCub and the target is only 3.5 meters, but there’s no reason it couldn’t be scaled up to larger distances. Or bigger arrows. Or rocket launchers.

This robot experiment was conducted by Dr. Petar Kormushev, Dr. Sylvain Calinon, and Dr. Ryo Saegusa at the Italian Institute of Technology (the same guys who brought you robot pancake flipping). You can read a bit more about it at the link below.

[ Robot Archer iCub ]

Thanks Dr. Kormushev!

iCub is a robot designed to study cognition and learning, and his latest talent is dynamic ball catching. Rather than being programmed to do this, iCub gets ‘taught’ by a human, who makes catching motions while being hooked up to some motion encoding hardware. This approach allows iCub to dynamically adapt to variable ball trajectories, which is the kind of thing that happens all of the time outside of the lab, as it were.

Obviously, iCub needs to speed up a bit if he wants to be useful in a baseball game, and he certainly doesn’t have anything on the speed or precision of robot hands like this or this. But, iCub also doesn’t depend on an array of high speed cameras, and he also doesn’t depend on a constant trajectory for the ball, making him far more adaptable. At this point, I’m not entirely sure if iCub needs faster hardware or software or both, but the potential is here for something pretty cool in the near future.

[ RobotCub iCub ]

iCub, who we’ve met before (a few times), was designed to study cognition in children. Thinking, learning, development, stuff like that. As such, iCub was physically modeled on a two year old. Back when iCub was first designed, though, the technology didn’t exist to make functional hands that small, so the robot was equipped with a set of hands equivalent in size to an eight year old. This has just been fixed, with iCub now sporting a pair ‘o mitts appropriate for its age. iCub also got a new, springier pair of legs that should be better able to manage the inevitable faceplants that happen when a child (or a robot child) is trying to teach itself how to walk.

A couple more iCub pics, including a disembodied head, after the jump. (Read more…)