Parameter optimisation of an evolutionary algorithm for on-line gait generation of quadruped robots

Abstract

This paper presents a hybrid evolutionary algorithm (EA) for developing locomotion gaits of Sony legged robots. An online training algorithm is used for generating gaits for quadruped walking robots based on a hybrid approach that changes the probability of genetic operators in respect to the performance of the operator's offspring. The probability of applying an operator changes in proportion to the observed performance of the individuals created by that operator in the course of a run. The selection of EA parameters such as the population size and recombination methods and mutation parameters are made to be flexible and strive towards optimal performance autonomously. An overhead CCD camera is used to evaluate the performance of the generated gaits on-line while the robot is playing a football game. Robot is learning to walk on its own without any human interference.