Policy Neural Control Barrier Function (PNCBF)

 

How to train your neural control barrier function: Learning safety filters for complex input-constrained systems

Oswin So, Zachary Serlin, Makai Mann, Jake Gonzales, Kwesi Rutledge, Nicholas Roy, Chuchu Fan

Webpage • arXiv • Paper   ❘   Installation • Getting started • Citation

Installation

This is a JAX-based project. To install, install jax first and other prereqs following their instructions. Note that the jax version used in this project is quite old (0.4.28). Next, clone the repository and install the package.

git clone https://github.com/mit-realm/pncbf.git
cd pncbf
pip install -e .

Getting started

Example on the double integrator:

python scripts/dbint/pncbf_dbint.py --name dbint

To eval,

python scripts/dbint/eval_pncbf_dbint.py runs/pncbf_dbint/path_to_run/ckpts/5000

Citation

Please cite the PNCBF paper.

@inproceedings{so2024train,
  title={How to train your neural control barrier function: Learning safety filters for complex input-constrained systems},
  author={So, Oswin and Serlin, Zachary and Mann, Makai and Gonzales, Jake and Rutledge, Kwesi and Roy, Nicholas and Fan, Chuchu},
  booktitle={2024 IEEE International Conference on Robotics and Automation (ICRA)},
  pages={11532--11539},
  year={2024},
  organization={IEEE}
}