mfnlc's Introduction

Neural Lyapunov Deep Reinforcement Learning

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

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

Install necessary dependencies.

pip install -e .

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

bash download.sh

Quick Start

Two quick start examples:

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

mfnlc's People

Contributors

Stargazers

Forkers

toshj

mfnlc's Issues

Does the repo use its own safetygym files?

Hi, thanks for the big contribution in safeRL field. There was mentioned that this repo doesnt contain safety_gym repo(also readme doesnt mention anything about it). So i downloaded it manually, but i have noticed files in
https://github.com/ZikangXiong/MFNLC/tree/master/mfnlc/envs

This is a modification of files from safety_gym repo(such as engine.py). Should i use these files in SafetyGym instead original one? What type of modification was made in these files, is it related only to env render, or maybe somehow change the robot models?

some issues about render and training

Hi, first of all, thank you for the quality of your work, which has inspired me.

In my study of your code, I have encountered a few problems.

First, when I was testing my environment using test_envs.py, the test_render function in env.render() was working with the prompt "render() got an unexpected keyword argument ' follow'", when I delete follow, vertical, scale, it renders but the robot does not move. (gym=0.26, mujoco=2.10, mujoco_py=2.1.2.14). I don't know if you have encountered this or if you can provide me with the relevant package version so that I can solve this bug

Then, when I try to train using the code of the exps species, I get "/data/shrl/e2e/Nav/model.zip.zip'", which is probably because I didn't find the relevant file, I wonder if you could provide

Thanks a lot.

Recommend Projects

zikangxiong / mfnlc Goto Github PK

mfnlc's Introduction

Neural Lyapunov Deep Reinforcement Learning

Project Structure

Install

Quick Start

Bibtex

mfnlc's People

Contributors

Stargazers

Forkers

mfnlc's Issues

Does the repo use its own safetygym files?

some issues about render and training

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent