verify-neural-CBF

This is the official code for CoRL 2024 work "Verification of Neural Control Barrier Functions with Symbolic Derivative Bounds Propagation".

Preparation

The code is based on Julia and is tested with Julia v1.9.4. Check here to install Julia environment. Install ModelVerification.jl from this repo and check out the branch verify_gradient here. Install RobotZoo.jl from this repo and TaylorModels.jl from here.

Data Collection

To collect data for each robot dynamics, see Jupyter file collect_data.ipynb for details.

Model training

For the model training under Point Robot, see Jupyter file train_naive_point.ipynb for regular training and train_adv_point.ipynb for adversarial training. For the model training under Dubins Car, see Jupyter file train_naive_car.ipynb for regular training and train_adv_car.ipynb for adversarial training. For the model training under Planar Quadrotor, see Jupyter file train_naive_planar_quad.ipynb for regular training and train_adv_planar_quad.ipynb for adversarial training.

Verification of neural CBFs

For the verificaiton under Dubins Car, see Jupyter file verify_car.ipynb. Similarly, verify_point.ipynb is for point robot and verify_planar_quad.ipynb is for planar quadrotor. Replace the corresponding path with naive or adv for different pre-trained models. Also, $\alpha$ and number of grids per dimension can also specified for ablation study. For baselines, specify max_iter=1 in BFS mehtod to specify NNCB-IBP and otherwise, it is for BBV baseline.

Citation

If you find the repo useful, please cite:

H. Hu, Y. Yang, T. Wei and C. Liu "Verification of Neural Control Barrier Functions with Symbolic Derivative Bounds Propagation", Conference on Robot Learning (CoRL). PMLR, 2024

@inproceedings{
hu2024verification,
title={Verification of Neural Control Barrier Functions with Symbolic Derivative Bounds Propagation},
author={Hanjiang Hu and Yujie Yang and Tianhao Wei and Changliu Liu},
booktitle={8th Annual Conference on Robot Learning},
year={2024},
url={https://openreview.net/forum?id=jnubz7wB2w}
}