Reinforcement learning for a snake-like robot controlled by a central pattern generator

Abstract

A snake has large potential to move in various environments by drastically changing its 'gait' pattern, in spite of its simple body. We configured a control scheme for a snakelike robot, using a central pattern generator, and developed a learning algorithm to acquire a good control rule in a changing environment. Although the snake-like robot has a large degree of freedom, a computer simulation showed that a control rule that allows the robot to move to a target direction on an inclined ground or in a changing environment is obtained by our scheme.