Abstract
In this study, we select a reinforcement learning technique among various means to make a robot adapt to variable or unknown environments. To evaluate the effects of components in a four-legged hopping robot on moving motion, we conducted a simulation experiment in which some virtual robots composed of different robotic components used trial and error to acquire adequate motions. We evaluated the effect of the existence of retractility of the legs on the motion of the robot, and further evaluated the effect of the existence of spring elements on the motion of the robot with retractility of the legs. As a result, the robots with spring elements and retractility of the legs finally acquired the motion with longest moving distance, because their components enhanced the strength for leaping and absorbed shock while landing.
