Learning-Guided Force-Feedback Model Predictive Control with Obstacle Avoidance for Robotic Deburring
Authors: Krzysztof Wojciechowski, Ege Gursoy, Arthur Haffemayer, Sebastien Kleff, Vincent Bonnet, Florent Lamiraux, Nicolas Mansard
Organizations: LAAS-CNRS, Universite de Toulouse, CNRS, Toulouse, France · Inria, AUCTUS team, Talence, France · Image and Pervasive Access Laboratory (IPAL), CNRS-UMI, 2955, Singapore · Artificial and Natural Intelligence Toulouse Institute, France
Model Predictive Control (MPC) is widely used for torque-controlled robots, but classical formulations often neglect real-time force feedback and struggle with contact-rich industrial tasks under collision constraints. Deburring in particular requires precise tool insertion, stable force regulation, and collision-free circular motions in challenging configurations, which exceeds the capability of standard MPC pipelines. We propose a framework that integrates force-feedback MPC with diffusion-based motion priors to address these challenges. The diffusion model serves as a memory of motion strategies, providing robust initialization and adaptation across multiple task instances, while MPC ensures safe execution with explicit force tracking, torque feasibility, and collision avoidance. We validate our approach on a torque-controlled manipulator performing industrial deburring tasks. Experiments demonstrate reliable tool insertion, accurate normal force tracking, and circular deburring motions even in hard-to-reach configurations and under obstacle constraints. To our knowledge, this is the first integration of diffusion motion priors with force-feedback MPC for collision-aware, contact-rich industrial tasks.
Figures & tables
Fig. 1 : Torque-controlled Franka Emika Panda robot executing the deburring task in scenario 3 (see sec. V-D ). The robot inserts a custom end-effector into different holes of the workpiece and applies a constant normal force while performing circular motions, all while adapting to varying joint configurations and avoiding collisions with the environment.
Fig. 2 : Robotic deburring process: the manipulator applies a constant normal force along the tool axis while performing a circular motion inside the hole to remove burrs with limited tangential forces.
Fig. 3 : Custom end-effector mount for the Franka Emika Panda integrating a force-torque sensor and provisions for RGB-Depth camera. A spherical tool tip enables smooth circular motion within holes. (Scenario 2, Sec. V-D )
Fig. 4 : Representative sequence of a circular deburring motion performed onto the object. (Scenario 2, Sec. V-D )
Fig. 5 : Scenario 1 (Sec. V-C ): Force-colored end-effector x-y trajectories for a single hole; color encodes normal normal force magnitude. (a) 1D contact, no obstacle avoidance; (b) 1D contact with an obstacle in front of the robot; (c) 3D contact, no collision avoidance; (d) 3D contact with collision avoidance. The obstacle shape is clearly visible in the successful example c, and prevents the robot to achieve the complete circle.
Fig. 6 : Scenario 1 (Sec. V-C ): Tracking of normal and tangential forces during deburring motion on a flat table with collision avoidance with 1D (top) and 3D contact (bottom) models. In this case when 1D contact model approaches collision, lateral forces exceed force friction cone ( μ=0.6 ) and break the contact while 3D contact maintains stable movement.
Fig. 7 : Positions of the 5 holes to be deburred on the investigated workpiece and reaching postures for each hole.
Hole
wf (–)
Force RMSE [N]
Hole 1
10.6956
11.30
Hole 2
6.4821
7.98
Hole 3
6.9166
17.20
Hole 4
19.0211
17.41
Hole 5
12.3884
15.34
TABLE I : Inverse velocity manipulability and corresponding force RMSE.
Fig. 8 : Description of the collision pair between the plane (red) and ellipsoid (yellow) (a). Force-colored end-effector x-y trajectories for a single hole. Color encodes normal force magnitude.
Department of Biomedical Engineering, National University of Singapore, Singapore 117583 · Peng Cheng Laboratory, Shenzhen 518057, China · School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 639798
Machines in Motion Laboratory, Center for Robotics and Embodied Intelligence, New York University (NYU), USA. · Artificial and Natural Intelligence Toulouse Institute (ANITI), France.
The Center for Humanoid Research, Korea Institute of Science and Technology (KIST), 02792 Seoul, South Korea · The School of Electrical Engineering, Korea University, 02841 Seoul, South Korea