Safe-MPC

This repo contains algorithms for Model Predictive Control, ensuring safety constraints and recursive feasibility.

Installation

• Clone the repository
  git clone https://github.com/idra-lab/safe-mpc.git
• Install the requirements
  pip install -r requirements.txt
• Inside the root folder, download the zip (containing the Neural Network weights) from here, rename it as nn_models.zip and unzip it.
• Follow the instructions to install CasADi, Acados and Pytorch.

Usage

Run the script main.py inside the scripts folder. One can consult the help for the available options:

cd scripts
python3 main.py --help

For example:

• find the initial static configurations of the manipulator on which we run the tests

python3 main.py -i

• obtain the initial guess (to warm-start the MPC), for a given controller

python3 main.py -g -c=receding

• run the MPC

python3 main.py --rti -c=receding

References

@misc{lunardi2023recedingconstraint,
      title={Receding-Constraint Model Predictive Control using a Learned Approximate Control-Invariant Set}, 
      author={Gianni Lunardi and Asia La Rocca and Matteo Saveriano and Andrea Del Prete},
      year={2023},
      eprint={2309.11124},
      archivePrefix={arXiv},
      primaryClass={cs.RO}
}