Adaptive control model of a locomotion by the central pattern generator

Most of basic locomotor patterns of living bodies are controlled by central pattern generators (CPGs) which are collective neural oscillators. The CPG sends control signals to muscular systems, and the activity of the CPG is strongly affected by sensory signals from the body. Therefore, it can be said that locomotor patterns are generated by the interaction between the CPGs and the body movements. To control a physical system, such as a leg, by the CPG, it would be necessary to obtain an adequate intrinsic frequency of the CPG and interactions between the CPG and the physical system. In this article, we regard a physical system as a physical oscillator and propose a learning algorithm to acquire these parameters aid apply the learning method to the control of a hopping robot. The proposed learning method does not need any information about the dynamics of the controlled physical system and requires only local informations like the Hebbian rule.