Soft computing approaches to motion control for humanoid robots

This paper introduces two topics of the soft computing approach to motion control for humanoid robots. The motion control primitive is based on central pattern generator and neuro-musculo-skeletal system. The paper covers the two types of robot: a two dimensional link model of entire body and a three dimensional link model specializing in lower body. The paper provides the first topic of an parallel stochastic method by two parallel processes, which handles the control parameters of upper-limbs and legs and communicates them each other, and the second topic of an phased learning method considering joint stiffness. The paper also shows the performance results of the walking motions generated by proposed methods.