Optimizationof CPG-network for decentralized control of a snake-like robot

Abstract

Periodic locomotion of animal bodies with large degree of freedom is known to be realized by decentralized control using network of central pattern generators (CPGs) that are distributed in spinal cord (in vertebrates) or nerve cords (in invertebrates). In this paper, optimization of a decentralized controller for a snake-like robot based on CPG-network is presented. The controller is composed of CPGs on every robot joints and connections between them. Oscillations of CPGs are coordinated by propagating them each other. For several types of network structure, genetic algorithm (GA) is used to optimize neural parameters, i.e. CPG parameters and synaptic weights. As a result, we derived a CPG-network that is optimal in terms of moving speed