London, August 7 : German researchers have written a piece of software that enables robots to "learn" to move through trial and error.
Ralf Der, a professor at the Max Planck Institute for Mathematics in the Sciences (MPI-MIS) in Leipzig, has revealed that the software mimics the interconnected sensing and processing of a brain in a so-called "neural network".
The researcher says that simulated creatures equipped with such a network start to explore-such as a robotic dog learning to jump over a fence, and a humanoid learning how to get upright and do back flips.
Prof. Der has also applied the software to simulated animals and humans.
Just in case a robot's neural network encounters an obstacle, such as a wall or the floor, it tries different moves, learning about itself and its environment as it does so.
"In the beginning, we just drop a robot into a space. But they don't know anything, so they don't do anything," the BBC quoted Prof. Der as saying.
Eventually, the neural network captures electronic noise that causes small motions, says the researcher.
Having learnt about its range of movement, the robot later tries larger motions.
"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," Prof. Der said.
The new software can be used with any kind of robot, and has been tried on simple wheeled systems.
"I call it a plug-and-play brain," Prof. Der said.
Daniel Polani of the University of Hertfordshire said: "The classic thing in robotics is 'bring this' or 'play this chess game and win'-the task is given. Ralf Der's system is only defined by what it perceives and does, but there's no goal. It's a very good approach."
Prof. Der, however, concedes that his system promptly forgets what it learns. His team is currently working to create a long-term memory.
He will make a presentation on his work at the Artificial Life XI conference in Winchester this week.