Model Predictive Control for Dynamic Cloth Manipulation: Parameter Learning and Experimental Validation

Robotic cloth manipulation is a challenging problem for robotic systems. Textile items can adopt multiple configurations and shapes during their manipulation. Hence, robots should not only understand the current configuration of the item but also be able to predict its future possible behaviors and perform real-time control during manipulation. This article addresses the problem of indirectly controlling the configuration of certain points of a textile object, by applying actions on other parts of it through the use of a model predictive control (MPC) strategy. MPC allows to foresee the behavior of indirectly controlled points, while satisfying physical/operational constraints. This is done by first identifying the optimal control signals that may constitute the desired future cloth configuration. After that, a dynamic model of the item will be used and sensor data will allow to update the belief on the object’s state and close the loop. This article investigates how grasping the upper corners of a square piece of cloth can allow to track a reference trajectory of the pieces’ lower corners. To do so, we propose and validate a linear cloth model that allows solving the MPC optimization problem in real time. Reinforcement learning (RL) techniques are used to learn the parameters of the proposed cloth model and to tune the resulting MPC. The full control scheme was implemented and executed in a real robot, obtaining accurate tracking results in adverse conditions.