In this paper, we propose a control method which is adaptive to the environment and apply it to a biped underactuated robot named ldquoKenkenIIrdquo which has springs mimicing a tendon system of animals so that KenkenII can walk and run stably from zero-velocity on irregular terrains. In the proposed method, we use a CPG(Central Pattern Generator) constructed by a neural oscillator network and realize rhythmic movements. Parameters in the control system are searched by PSO(Particle Swarm Optimization) and the validity is conformed by numerical simulation and experiment.