PARC presents a parallelizable motion-planning framework for the reach-avoid problem in autonomous robotics, guaranteeing both safety and liveness. It models the planning dynamics as a parameterized piecewise-affine system, allowing for the computation of a continuum of feasible motion plans through a precise backward reachability analysis. The utility of PARC is demonstrated across various robotics systems, where it is benchmarked against leading reach-avoid methods, showcasing its superiority in low conservatism.

[Homepage]   [Study Paper]   [Safe Robotics Lab @ GT]

Authors: Long Kiu Chung* ([email protected]), Wonsuhk Jung* ([email protected]), Chuizheng Kong ([email protected]), and Shreyas Kousik ([email protected]).

*Equal Contribution

• [03/11/2024] v0.1.0: Initial code and paper release

To run this code, you will need

1. MPT3 Toolbox
2. simulator (Download and add to your path)

To reproduce the comparison results shown in the paper, please navigate to the demo/system_name/benchmark (e.g., demo/nearhover/benchmark) folder.

1. For the comparison with FasTrack, we provide a pre-computed reachable set here using Level-set Toolbox.
2. For the comparison with NeuralCLBF, we provide a pre-trained weight here and corresponding simulated result here.
3. For the comparison with RTD-quadrotor, we provide a pre-computed forward reachable set here.
4. To generate the comparison plots in the paper, please install CORA2018.

demo folder provides you a step-by-step guide how to write your own PARC algorithm via learning example of turtlebot 2D navigation and nearhover 3D navigation.

1. agents: Abstraction for the tracking model (e.g., Turtlebot)
2. TrajectoryModel: Abstraction for the planning model (e.g., Dubins Car)
3. utils: Collection of useful operations as visualization, set operation, sampling, that is agnostic to choice of tracking model and planning model.

Please cite this paper if you use PARC in an academic work:

@article{chung2024goal,
  title={Goal-Reaching Trajectory Design Near Danger with Piecewise Affine Reach-avoid Computation},
  author={Chung, Long Kiu and Jung, Wonsuhk and Kong, Chuizheng and Kousik, Shreyas},
  journal={arXiv preprint arXiv:2402.15604},
  year={2024}
}