Formative behavior network for a biped robot; a control system in consideration of motor development

The purpose of this paper is to propose a control system which enables a biped robot to walk sideways. The control system, named Formative Behavior Network (FBN), has a structure which consists of movement elements responding to sensory stimuli. We call each element a 'Behavior.' We do not arrange the number of Behaviors nor their parameters beforehand. With appropriate rewards for actions of the robot, the FBN is able to learn to achieve the task even if no Behavior exists at first. In the simulations, the robot implemented with the FBN learned to walk and showed its adaptability to a change of the robot model, such as lengthening its legs. Based on these results, an application to the experimental biped robot also succeeded in the real environment.