Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : June 28, 2023 - July 01, 2023
In this paper, we consider a throwing manipulation using a 1-DOF robot to manipulate an object on a robot arm to a target position through a throwing motion in a 2D plane. In a previous study, we showed that Bayesian optimization could improve throwing accuracy against stochastic uncertainty. However, the object’s initial position was constrained to the arm end, and the generalization performance was insufficient. In this study, we investigated a motion generation method that can throw an object placed at an arbitrary position on an arm to a target position, with the aim of improving generalization performance. In particular, we studied a motion generation method based on reinforcement learning and verified its effectiveness through simulation experiments.
