Motion Control System with Emergent Learning Ability and Simulation of Gait Control.

摘要

The purpose of this research is to propose and develop a motion control method which has adaptive and emergent ability such as biological system. In developing a motion controller, it becomes a hindrance that parameters of robots undergo changes as time passes. Muscular systems have much nonlinear properties and those parameters often change, however, motion controller of biological system can handle them properly. It is to be desired that motion pattern should be changed according to motion parameter changes. So, we try to develop these types of motion controller by use of RNN (Recurrent Neural Network) and GA (Genetic Algorithm) . Parameters in the RNN are treated as the genetic information. Time constants, synaptic weights and thresholds of neurons are crossed over and mutated and suitable motion patterns would be obtained according to the performance indices. We tried to learn and control locomotion of biped robot which parameters are unknown. As a result of simulations, the motion controller acquired several patterns of gaits. Therefore, it is developed that, the motion controller which can learn and generate motion patterns as to parameters of motion system and changes of environments.