A Learning Method of Dual-arm Manipulation for Cloth Folding Using Physics Simulator

Abstract

This paper describes a learning method of cloth folding manipulation by a dual-arm robot. To reduce the burden on doing manipulation experiments using actual robots, cloth manipulation is virtually learned using physics simulator. The issue on using virtual environments is how to reproduce realistic movements of cloth in the simulator. We adopt methods of determining simulation parameters from three kinds of actual measurement. In addition, we propose a method using Bayesian optimization in order to obtain folding manipulation efficiently using the simulator. The method enables to select appropriate gripping positions and moving trajectories. The effectiveness of these methods were confirmed by means of simulation, and also by experiments using an actual robot.