Skip to content
/ TD3_BC Public

Author's PyTorch implementation of TD3+BC, a simple variant of TD3 for offline RL

License

Notifications You must be signed in to change notification settings

sfujim/TD3_BC

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

A Minimalist Approach to Offline Reinforcement Learning

TD3+BC is a simple approach to offline RL where only two changes are made to TD3: (1) a weighted behavior cloning loss is added to the policy update and (2) the states are normalized. Unlike competing methods there are no changes to architecture or underlying hyperparameters. The paper can be found here.

Usage

Paper results were collected with MuJoCo 1.50 (and mujoco-py 1.50.1.1) in OpenAI gym 0.17.0 with the D4RL datasets. Networks are trained using PyTorch 1.4.0 and Python 3.6.

The paper results can be reproduced by running:

./run_experiments.sh

Bibtex

@inproceedings{fujimoto2021minimalist,
	title={A Minimalist Approach to Offline Reinforcement Learning},
	author={Scott Fujimoto and Shixiang Shane Gu},
	booktitle={Thirty-Fifth Conference on Neural Information Processing Systems},
	year={2021},
}

*This is not an official Google product.

About

Author's PyTorch implementation of TD3+BC, a simple variant of TD3 for offline RL

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published