gym_multirotor

Gym to train reinforcement learning agents on UAV platforms

Quadrotor	Tiltrotor

Requirements

• This package has been tested on Ubuntu 18.04/20.04 with python 3.8.
• To install MuJoCo binaries refer this.
• Few additional packages:

  pip install numpy scipy
  pip install mujoco_py==2.1.2.14
  pip install stable-baselines3[extra]
  pip install gym==0.21.0
  

• For troubleshooting refer this
• To install gym refer this link.

Installation

To install, you will have to clone this repository on your personal machine. Follow the below commands:

$ git clone https://github.com/adipandas/gym_multirotor.git
$ cd gym_multirotor
$ pip install -e .

Environments

List of environments available in this repository include:

Environment-ID	Description
QuadrotorPlusHoverEnv-v0	Quadrotor with + configuration with task to hover.
TiltrotorPlus8DofHoverEnv-v0	Tiltrotor with + configuration.
QuadrotorXHoverEnv-v0	Quadrotor with x configuration with a task to hover.

How to use?

Please refer examples folder

References

REFERENCES.md

Citation

If you find this work useful, please cite our works:

@inproceedings{deshpande2020developmental,
  title={Developmental reinforcement learning of control policy of a quadcopter UAV with thrust vectoring rotors},
  author={Deshpande, Aditya M and Kumar, Rumit and Minai, Ali A and Kumar, Manish},
  booktitle={Dynamic Systems and Control Conference},
  volume={84287},
  pages={V002T36A011},
  year={2020},
  organization={American Society of Mechanical Engineers}
}

@article{deshpande202190Robust,
title = {Robust Deep Reinforcement Learning for Quadcopter Control},
journal = {IFAC-PapersOnLine},
volume = {54},
number = {20},
pages = {90-95},
year = {2021},
note = {Modeling, Estimation and Control Conference MECC 2021},
issn = {2405-8963},
doi = {https://doi.org/10.1016/j.ifacol.2021.11.158},
url = {https://www.sciencedirect.com/science/article/pii/S2405896321022023},
author = {Aditya M. Deshpande and Ali A. Minai and Manish Kumar}
}

Notes:

• Some of the environment parameters have been updated but the task of these drone environments still remains the same as what was discussed in the paper.
• I will keep on updating these codes as I make further progress in my work.