Evolution strategies for biped locomotion learning using nonlinear oscillators

This paper addresses a tuning method for the parameters in a gait pattern, control systems and nonlinear oscillators of humanoid robots. The phase oscillators with PD controller are used to generate a rhythmic walking pattern, thus a walking pattern is described by many parameters. Using an evolutionary computation, our approach uses only performance evaluation values to tune these parameters. In this paper, we propose a novel evolution strategy that uses mask on the portion of individual to avoid mutation. Numerical simulation studies are carried out to evaluate the performance of the proposed approach by using the RoboCup 3D Soccer Simulator.