Researchers in Leipzig have invented software that allegedly allows robots to learn movement through trial and error.
The software mimics the interconnected sensing and processing of a brain via a "neural network," allowing the simulated creatures to explore.
For instance, the video demonstrations, a simulated dog learns to jump over a fence, and a humanoid learns how to get upright and do back flips. The key word here being “simulated,” as they haven’t actually made any of this happen yet. In fact, the video demo of the software looks like it was created in 1983…but I digress…
So how does it work, you wonder? Well, in the case of a simulated human, there are 15 joints and the angles through which they can move. The network sends out signals to move in a specific way, and predicts where the human-bot should end up, based on that movement. If it encounters an obstacle, the the robot tries different moves , learning about itself and its environment in the process.
"In the beginning, we just drop a robot into a space. But they don't know anything, so they don't do anything," Professor Der of Leipzig said. "The neural network eventually picks up on electronic noise, which causes small motions. Soon it tries larger motions and learns about its range of movement within that environment. It's like a newborn baby—it doesn't know anything but tries motions that are natural for its body. Half an hour later, it's rolling and jumping," Professor Der said. “I call it a plug-and-play brain.”
Dr. Der will present the video demonstrations at the Artificial Life XI conference in Winchester this week.
BBC.co.uk: Robots learn to move themselves