Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : December 09, 2021 - December 10, 2021
The function of robot manipulation can be extended by attaching tools to end-effector. In addition, deep imitation learning can be used to make the robot arm imitate the use of tools. However, deep learning requires a large number of iterations. Therefore, It is not suitable in actual clearing scene. In this paper, a method to reduce the number of training iterations is proposed by loading deep learning parameters that represent the usage of another tools as initial values. In this paper, the reduced number of iterations and the effectiveness of the method are confirmed. A cleaning experiment is also conducted with silica sand.
