Force Feedback MPC

Optimal control toolbox to achieve force feedback control in MPC. This library is basically an extension of the Crocoddyl optimal control library: it implements custom action models in C++ with unittests and python bindings. In particular, it contains the core classes used in MPC experiments of the following papers:

• S. Kleff, et. al, "Introducing Force Feedback in Model-Predictive Control", IROS 2022. PDF
• S. Kleff, et. al, "Force Feedback in Model-Predictive Control: A Soft Contact Approach" PDF (under review)

The code to reproduce our experiments (i.e. real-time implementation of the force-feedback MPC), along with our experimental data are available in this separate repository.

Dependencies

Python compatibility: 3.10-3.13

Core dependencies

Package	Version	Purpose
Pinocchio	3.8.0	Robot dynamics & kinematics
Crocoddyl	>= 3.2.0	Optimal control library
mim-solvers	0.2.0	Optimization solvers
CMake	>= 3.10	Build system

Demo dependencies

For force tracking & polishing demos (Kuka Iiwa):

• NumPy, SciPy, Matplotlib
• PyBullet (physics simulation)

For Go2 multi-contact demos (requires above + specific deps below):

• MuJoCo (physics engine)
• meshcat-python (3D visualization)
• OpenCV (computer vision - Go2Py dependency)
• Go2Py (Go2 quadruped interface)

Installation

Using the provided conda environment file:

# 1. Create environment from file
conda env create -f environments/force_feedback_mpc.yml
conda activate force_feedback_mpc

# 2. Build and install force_feedback_mpc from source
mkdir build && cd build
cmake .. -DCMAKE_INSTALL_PREFIX=$CONDA_PREFIX
make -j4
make install

# 3. Install required Python packages (from workspace root)
cd ..
pip install -e ./croco_mpc_utils --no-deps 
pip install -e ./mim_robots --no-deps

# 4. [Optional] Install Go2Py for Go2 demos
pip install -e /path/to/Go2Py

Running the Demos

The demos are located in the demos/ directory. First, activate the environment:

conda activate force_feedback_mpc

Classical MPC (rigid contact force model)

python demos/force_tracking/classical/force_tracking_classical_mpc.py

Force-feedback MPC (soft contact force model)

python demos/force_tracking/soft/force_tracking_soft_mpc.py

Polishing Task (apply constant normal force while tracking end-effector circle)

python demos/polishing/classical/polishing_classical_mpc.py
python demos/polishing/soft/polishing_soft_mpc.py

Go2 Quadruped (whole-body multi-contact MPC)

Requires Go2Py to be installed.

python demos/go2arm/Go2MPC_demo_classical.py
python demos/go2arm/Go2MPC_demo_soft.py

Citing this work

@unpublished{kleff:hal-04572399,
  TITLE = {{Force Feedback in Model-Predictive Control: A Soft Contact Approach}},
  AUTHOR = {Kleff, S{\'e}bastien and Jordana, Armand and Khorrambakht, Rooholla and Mansard, Nicolas and Righetti, Ludovic},
  URL = {https://hal.science/hal-04572399},
  NOTE = {working paper or preprint},
  HAL_LOCAL_REFERENCE = {Rapport LAAS n{\textdegree} 24093},
  YEAR = {2025},
  MONTH = Jun,
  PDF = {https://hal.science/hal-04572399v2/file/force_feedback_article_second_submission.pdf},
  HAL_ID = {hal-04572399},
  HAL_VERSION = {v2},
}