Flattening and folding towels with a single-arm robot based on reinforcement learning

Abstract

Robots can learn how to complete a variety of tasks without explicit instructions thanks to reinforcement learning. In this work, a piece of cloth is placed on a table and manipulated using a single-arm robot. We consider 2 forms of manipulation: flattening a crumpled towel and folding a flat one. To learn a policy that will allow the robot to select the optimum course of action based on observations of the environment, we construct a simulation environment using a gripper and a piece of cloth. After that, the policy is applied to a real robot and put to the test. Additionally, we present our method for identifying the corners of a garment using computer vision, which includes a comparison between a traditional computer vision approach with a deep learning one. We use an ABB robot and a 2D camera for the experiments and PyBullet software for the simulation.