Skip to: Content
Skip to: Site Navigation
Skip to: Search


Horizons

Computers “learn” to walk

By / August 7, 2008



A boxy humanoid wiggles on the ground. Like a child taking his first steps, it plants its blue shoes and wobbles toward standing up. Then its balance crumbles and the on-screen computer character plops down to try again.

Skip to next paragraph

Recent posts

German researchers designed this simulator to allow their little blue man to “learn” movements. Rather than program the digital character to perform specific motions, the code allows for trial and error – an important step in creating fully functioning robots. Once it masters moving in a virtual world, scientists can translate it to the real one.

As Northeastern University biology professor Joseph Ayers told the Monitor in June: “If you’re controlling a robot with a computer program, unless you’ve anticipated every possible situation it’s going to get into, it will eventually get into a situation where it has no escape strategy and it will be stuck. Animals never get stuck. What animals do is they wiggle and squirm [until they escape].”

This new German simulator attempts to inject that natural spirit into robotics. While still trapped in a computer, the creation is free to wander the virtual realm – it just needs to figure out how to get there. (Check out a video clip here.)

Bound only by 15 joints and with caps on how far each will bend, the humanoid freely flails through potential movements in an attempt to accomplish a set goal. Sometimes it’s told to stand upright, sometimes to compete a back flip.

"It's like a newborn baby,” Professor Der told the BBC. “It doesn't know anything but tries motions that are natural for its body. Half an hour later, it's rolling and jumping.”

Der’s team has also played around with simulated dogs and a pork-bellied creature that bops about like a bucking bronco.

This early in the experiment, the creations learn their way around and then quickly forget – allowing Der to monitor how they conquer obstacles the second or third time around. But the team is designing a memory program, so the ragdolls can build up successful routines and apply them to new challenges.

[Via the BBC]

Read Comments

View reader comments | Comment on this story