LeapLabTHU / FamO2O

Repository of "Train Once, Get a Family: State-Adaptive Balances for Offline-to-Online Reinforcement Learning" (NeurIPS 2023 Spotlight)

[NeurIPS 2023 Spotlight] Train Once, Get a Family: State-Adaptive Balances for Offline-to-Online Reinforcement Learning

This repository is the official source code for Train Once, Get a Family: State-Adaptive Balances for Offline-to-Online Reinforcement Learning [arXiv page] [project page] [OpenReview page], which has been accepted as a spotlight presentation at NeurIPS 2023. (Primary Contact: Shenzhi Wang)

This codebase includes:

1. The implementation of FamO2O using JAX IQL, located in the jax_iql folder. For detailed instructions, please see the jax_iql README.
2. The implementation of FamO2O using JAX CQL, located in the jax_cql folder. For additional information, please refer to the jax_cql README.

We would greatly appreciate it if you could cite our work!

@inproceedings{
wang2023train,
title={Train Once, Get a Family: State-Adaptive Balances for Offline-to-Online Reinforcement Learning},
author={Shenzhi Wang and Qisen Yang and Jiawei Gao and Matthieu Gaetan Lin and Hao Chen and Liwei Wu and Ning Jia and Shiji Song and Gao Huang},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023},
url={https://openreview.net/forum?id=vtoY8qJjTR}
}