1A2-K15 Adaptation of Leg/Wheeled Autonomous Rover by Reinforcement Learning(Space Robotics and Mecahtoronics)

An autonomous rover that is used as planetary exploration or rescue should work in unknown environment. Therefore, the rover needs to adapt to a new environment and should have to change its moving form. Therefore, a rover needs to judge situations around it, and should select the best action autonomously. This study aims at creating a rover which gets the best motion strategy for the environment or the action to recover from a stacked situation autonomously by using legs and wheels. For the adaptation method, we use a Reinforcement Learning scheme. To investigate availability of rover and to see the effect of learning, this study prepared virtual situations to escape from irregular ground as the experiments. As a result, the effectiveness of the method was shown. However, the result shows that the parameter largely effects over learning speed and the quality of solutions.