OSR

Code for NeurIPS 2023 paper "Recovering from Out-of-sample States via Inverse Dynamics in Offline Reinforcement Learning".

We have released the OSR demo v0.01.

Installation

1. If you want to install [OSR] environment, you can use the included Ananconda environment

   $ conda env create -f environment.yml
   $ source activate [OSR]

Run Experiments

You can run OSR experiments using the following command:

python -m conservative_sac_main \
    --env [ENVIRONMENT]

If you want to run on CPU only, just add the --device='cpu' option.

Credits

The CQL implementation is based on CQL. Thanks for their sharing codes.

Code for OSR-v and OSR-10 is coming soon...