1A2-C08 Method of Acquiring Motion of Gecko-Type Four-Legged Robot with Waist Joint Based on Reinforcement Learning(Evolution and Learning for Robotics(2))

Abstract

In this study, we focus on reinforcement learning as a method of acquiring a robot's motion autonomously and with adjustment to the environment. To acquire various motions with the learning method, we suggested a method of learning to gain new motion by combining one motion with another. In the experiment, we applied control regulation with this method of learning to the robot; then, after the robot learned both moving-forward and turning motions, we had the robot acquire motion to go straight in the target direction. As a result, the robot learned the new motion, and at the same time improved the other two motions.