Pytorch-based HJ Reachability Computation Toolbox.

Source code of "Data-Driven Hamiltonian for Direct Construction of Safe Set from Trajectory Data", IEEE CDC 2025.

Installation

Python version: 3.12

pip install -r requirements.txt

pip install -e .

Quick navigation

• dynamics.py: Dynamics & Optimal Control
• solver.py: Functions for various reachability problems and HJ-PDE numerical solution.
• grid.py: Grid class in Numpy and Torch.
• vis.py: Helper functions for visualizations.
• eval.py: Helper functions for simulation.
• experiment_design.py: Safe set expansion algorithm.

Examples

Model-based Computation:

• scripts/random_poly_2d/test_random_poly_2d_main.py
• scripts/xv15_3d/XV15_3d_in_vgamma_reachavoid.py

Data-Driven Hamiltonian:

• scripts/random_poly_2d/test_random_poly_2d_datadriven.py
• scripts/xv15_3d/XV15_3d_in_vgamma_expdesign_datareduction.py

Citation

@InProceedings{choi2025,
author={Choi, Jason J and Strong, Christopher A and Sreenath, Koushil and Cho, Namhoon and Tomlin, Claire J},
booktitle    = {IEEE Conference on Decision and Control (CDC)},
title        = {Data-Driven Hamiltonian for Direct Construction of Safe Set from Trajectory Data},
year         = {2025}
}