Motion generation for bipedal robot using neuro-musculo-skeletal model and Simulated Annealing

We propose a stochastic approach to CPG-based motion generation of a bipedal robot. We make an extension to a motion control system based on a neuro-musculoskeletal model into three dimensions, and give an optimization of control parameters by simulated annealing. Not only do we obtain walking motion by our approach but provide some empirical verification of various walking patterns adaptive to different link structures. In addition to straight walking, this paper gives the performance results of turning walk to right and left.