Caterpillar robot locomotion based on Q-Learning using objective/subjective reward

This paper presents an application of reinforcement learning, an unsupervised learning method, to a biological robot. This study focused on the primitive forward motion of a caterpillar robot to reveal how the robot obtains an optimal motion form in the Q-Learning process. First, this paper verifies that Q-Learning allows the caterpillar robot to move in the forward direction and examines the evolutionary process. Next, this paper discusses the emergence of the motion form using objective rewards. Finally, it examines the emergence of the motion using Q-Learning with subjective rewards in order to clarify the difference between the learning results. This examination provides a novel perspective on human robot interaction (HRI) via reinforcement learning.