NeuralPARC (ICRA'25) is an extension of PARC (RSS'24), which computes a set of initial positions and trajectory parameters with which a robot is guaranteed to reach a goal through narrow gaps. It improves upon the original method in three ways:

1. No assumption on system dynamics (black-box);
2. No need for a nominal plan;
3. Can arbitrarily decrease modelling error by adding more neurons.

Since NeuralPARC is data-driven, it is agnostic to how the black-box trajectories are generated as long as they are well-behaved and modellable, enabling it to provide safety guarantees on top of extreme maneuvers (e.g. drifting) or even learning-based policies (e.g. boatsim).

[Website]   [Paper]   [Video]

Authors: Long Kiu Chung, Wonsuhk Jung, Srivatsank Pullabhotla, Parth Shinde, Yadu Sunil, Saihari Kota, Luis Felipe Wolf Batista, Cédric Pradalier, and Shreyas Kousik.

• [2024/10/02] v0.1.0: Initial code release
• [2025/03/03] v0.1.1: Updated citations

To run this code, you will need

1. MPT3 Toolbox
2. simulator

1. Declare init_drifting or init_boatsim within the files in main to see NeuralPARC in action.
2. For drifting, we provide scripts for rolling out trajectories, validation, and visualization.
3. For boatsim, trajectories are collected and validated in Isaac Sim. We provide a subset of the collected trajectories in boatsim_rawdata.mat so the rest of the pipeline could be run on this repo.

To run NeuralPARC on a custom dynamical system, simply provide _rawdata.mat and init files with the same structure as the provided examples. Then run the files in main to apply NeuralPARC.

Please cite this paper if you use NeuralPARC:

@article{chung2025guaranteed, title={Guaranteed Reach-Avoid for Black-Box Systems through Narrow Gaps via Neural Network Reachability}, author={Chung, Long Kiu and Jung, Wonsuhk and Pullabhotla, Srivatsank and Shinde, Parth and Sunil, Yadu and Kota, Saihari and Batista, Luis Felipe Wolf and Pradalier, C{\'e}dric and Kousik, Shreyas}, booktitle={2025 IEEE International Conference on Robotics and Automation (ICRA)}, year={2025}, organization={IEEE} }