Home

Awesome

DrQ-v2: Improved Data-Augmented RL Agent

This is an original PyTorch implementation of DrQ-v2 from

[Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning] by

Denis Yarats, Rob Fergus, Alessandro Lazaric, and Lerrel Pinto.

<p align="center"> <img width="19.5%" src="https://i.imgur.com/NzY7Pyv.gif"> <img width="19.5%" src="https://imgur.com/O5Va3NY.gif"> <img width="19.5%" src="https://imgur.com/PCOR9Mm.gif"> <img width="19.5%" src="https://imgur.com/H0ab6tz.gif"> <img width="19.5%" src="https://imgur.com/sDGgRos.gif"> <img width="19.5%" src="https://imgur.com/gj3qo1X.gif"> <img width="19.5%" src="https://imgur.com/FFzRwFt.gif"> <img width="19.5%" src="https://imgur.com/W5BKyRL.gif"> <img width="19.5%" src="https://imgur.com/qwOGfRQ.gif"> <img width="19.5%" src="https://imgur.com/Uubf00R.gif"> </p>

Method

DrQ-v2 is a model-free off-policy algorithm for image-based continuous control. DrQ-v2 builds on DrQ, an actor-critic approach that uses data augmentation to learn directly from pixels. We introduce several improvements including:

<p align="center"> <img src="https://i.imgur.com/SemY10G.png" width="100%"/> </p>

These changes allow us to significantly improve sample efficiency and wall-clock training time on a set of challenging tasks from the DeepMind Control Suite compared to prior methods. Furthermore, DrQ-v2 is able to solve complex humanoid locomotion tasks directly from pixel observations, previously unattained by model-free RL.

<p align="center"> <img width="100%" src="https://imgur.com/mrS4fFA.png"> <img width="100%" src="https://imgur.com/pPd1ks6.png"> </p>

Citation

If you use this repo in your research, please consider citing the paper as follows:

@article{yarats2021drqv2,
  title={Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning},
  author={Denis Yarats and Rob Fergus and Alessandro Lazaric and Lerrel Pinto},
  journal={arXiv preprint arXiv:2107.09645},
  year={2021}
}

Please also cite our original paper:

@inproceedings{yarats2021image,
  title={Image Augmentation Is All You Need: Regularizing Deep Reinforcement Learning from Pixels},
  author={Denis Yarats and Ilya Kostrikov and Rob Fergus},
  booktitle={International Conference on Learning Representations},
  year={2021},
  url={https://openreview.net/forum?id=GY6-6sTvGaf}
}

Instructions

Install MuJoCo if it is not already the case:

Install the following libraries:

sudo apt update
sudo apt install libosmesa6-dev libgl1-mesa-glx libglfw3

Install dependencies:

conda env create -f conda_env.yml
conda activate drqv2

Train the agent:

python train.py task=quadruped_walk

Monitor results:

tensorboard --logdir exp_local

License

The majority of DrQ-v2 is licensed under the MIT license, however portions of the project are available under separate license terms: DeepMind is licensed under the Apache 2.0 license.