Research Project

MP-CloL: Learning Robotic Cloth Manipulation based on Physics Models and Model Predictive Control


UPC Project

Start Date


End Date


Project Code


Default project illustration


Project Description

This is a project within the Call for IRI MdM Internal Projects 2020 a call of R&D projects addressed to IRI early career researchers, under the María de Maeztu Unit of Excellence Programme.

The Institut de Robòtica i Informàtica Industrial has been focusing on making progress on cloth manipulation and assistive robotics during the last years. In some applications, Dimensionality Reduction (DR) techniques with motion characterization and reinforcement learning was successful at learning to fold a polo cloth. However, we realized how sensitive the ouput of the action is to any perturbation on the initial conditions, resulting in large noise in building the mapping from a robot motion parametrization to a reward function to optimize. This hinted the need to include some kind of predictive behaviour and prior knowledge in cloth manipulation.
Therefore, the natural next step is to not only apply compliant control methods, but also to use computationally-efficient cloth models in order to anticipate the cloth behaviours and have the robot proactively modifying its motion/torque accordingly. Integrating cloth models and control with reinforcement learning and propioceptive feedback is a multidisciplinary task that requires collaboration between IRI’s research groups.
This project aims at hiring one or two highly-motivated master students to implement and integrate such methods on the Barrett’s WAM robots, so that we can control a manipulated cloth’s behaviour by controlling the robot grasping it. The student will be supervised by Dr. Adrià Colomé (the PI of this project), regarding the robotics part, and Prof. Carlos Ocampo- Martinez, giving his advise on MPC.