Neural Lyapunov Deep Reinforcement Learning

Code Repository of IROS 22' paper Model-free Neural Lyapunov Control for Safe Robot Navigation

ArXiv | Demos

demo.mp4

Project Structure

├── README.md
├── setup.py
└── shrl
    ├── config.py       # config file, including data path, default devices, ect. 
    ├── envs            # simulation environments
    ├── evaluation      # evaluation utils
    ├── exps            # experiment scripts
    ├── learn           # low-level controller and neural Lyapunov function learning algorithms
    ├── monitor         # high-level monitor
    ├── plan            # high-level planner, RRT & RRT*
    └── tests           # test cases

Install

1. Install necessary dependencies.

pip install -e .

2. Configure MuJoCo-py by following official README.
3. (Optional) Download pretrained models (~35 MB)

bash download.sh

Quick Start

Two quick start examples:

1. Co-learning low-level controller and neural Lyapunov function
   python exps/train/no_obstacle/lyapunov_td3/[robot-name].py

2. Pre-compute monitor and evaluate
   python exps/hierachical/rrt_lyapunov/[robot-name].py

One can start tracing code from exps folder.

Bibtex

@inproceedings{Xiong2022ModelfreeNL,
  title={Model-free Neural Lyapunov Control for Safe Robot Navigation},
  author={Zikang Xiong and Joe Eappen and Ahmed H. Qureshi and Suresh Jagannathan},
  booktitle={2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  year={2022},
}