UCSD-SASLab/HJ-Patch

Patching Neural Barrier Functions with HJ Reachability

This repository contains the implementation of HJ-Patch accompanying the paper Patching Neural Barrier Functions using Hamilton-Jacobi Reachability by Sander Tonkens, Alex Toofanian, Zhizhen Qin, Sicun Gao, and Sylvia Herbert, submitted to IEEE Conference on Decision and Control (CDC), 2023

This project locally patches almost-barrier functions to guarantee safety, providing a 10-100x speedup over using vanilla HJ reachability

Requirements

• hj_reachability package: https://github.com/toofanian/hj_reachability. Custom fork that accepts an ``active'' region to be updated (although no speed up)
• optimized_dp package: https://github.com/toofanian/optimized_dp. Custom fork that accepts an ``active'' region and has a speedup from only computing on subset of states
• cbf_opt package: https://github.com/stonkens/cbf_opt
• For post-hoc analysis: experiment_wrapper package: https://github.com/stonkens/experiment_wrapper

Installation

Install all requirements and its dependencies using pip install -e . in your local conda environment. Then clone this repository and run pip install -e .

Usage

• refineNCBF contains all source code for interfacing with the optimized_dp and the hj_reachability solvers. Contains the HJ-Patch and vanilla HJ reachability implementations and a wide variety of implementation possibilities (different expansion methods, different breaking conditions, etc.)
• scripts: Contains codes that were used for generating the results in the paper
• data: Contains all the data functions (lfs will be added at a later stage)