Robots learn to move himself

Researchers showed in Leipzig software which is used for robots that allows them to learn to move through trial and error.

The software simulates association detection and prosecution of the brain in so-called “neural networks”.

Armed with this network, simulations to study the creatures.

During the demonstration video simulation dog learns to jump over a fence and humanoid learns to maintain and do the tour again

Newborn babies

Ralph at the Max Planck Institute for Mathematics in the Natural Sciences is also modeling software animals and humans.

Only the entrance to the kind of network traffic, the robot can reach in the case of humanoid, there are 15 teams, as well as the angle may be postponed. No information about the situation robot.

The network sends signals to move in a certain sense, and predicts that they should stop, on the basis of this motion.

If you feel obstacle, like himself, wall or floor, a prediction wrong, and the robot moves otherwise trying to learn more about himself and his environment, how it works.

“In the beginning, we only drop of robots in space. But they do not know anything that you do nothing,” said the professor. Neural networks finally reflected in the electronic noise of small movements.

This is ultimately a more comprehensive proposals, as you learn more about your range of motion. “It’s like a newborn baby knows nothing, but tried to natural body movements. Polchasa later, is a renewable and jumping,” said the professor.

Autonomous

This approach is much more flexible than traditional programs, in which the movements are carefully planned, well in certain places. In the face of change, so can the behaviour of the robot.

In addition, the software for each type of robot, and Professor He tried the system in simple systems wheels. “I call it a plug and play” brain “, he said.

“The classic thing in robotics, it brings” or “play the game of chess to win,” and challenge “, said Daniel Poland at the University of Hertfordshire. “Ralph only determined that he sees and makes But there is no purpose. This is a very good approach.”

At present, the network learned behavior, such as up, but quickly forget about them. And his colleagues are working on a long-term memory, so that when the robot is in a similar situation, we know what to do.

Video demonstrations XI artificial life conference in Winchester this week.

adapted from http://news.bbc.co.uk/