アーム搭載移動ロボットの駆動系故障時のための強化学習を用いたリカバリモーション獲得

概要

This study presents a new approach to obtain recovery motion for an arm mounted teleoperated crawler robot in drive system’s failure. A robot in drive system’s fault and in dangerous areas such as disaster sites needs to move. When one side crawler mechanism is in fault, the robot can use redundancies such as its arm for moving. It calls recovery motion in this study. However, it is difficult to know how to leverage these redundancies and to manipulate the robot. Our approach uses the reinforcement learning which makes robot do trial and error to maximize total reward it receives and finds the motion of purpose. To obtain the recovery motion, in advance by using reinforcement learning in the 3D dynamics simulator is effective for real robots. For reinforcement learning, three types of reward functions are used. In the 3D simulator, experiments on a crawler robot verified this approach.