This github organization contains implementations for our REPPO algorithm in several different popular RL frameworks, and we are adding more!
To read more about REPPO, please visit our webpage, or read the paper.
For the main paper code, aranged as easy-to-use jax single file implementations, see REPPO - Main.
For a torch implementation in the style of stable-baselines, check out our contrib fork.
Finally, for easy integration with robotics projects built on RSL-RL, check out our RSL-RL version.
Please cite REPPO as
@inproceedings{
voelcker2026relative,
title={Relative Entropy Pathwise Policy Optimization},
author={Claas A Voelcker and Axel Brunnbauer and Marcel Hussing and Michal Nauman and Pieter Abbeel and Radu Grosu and Eric Eaton and Amir-massoud Farahmand and Igor Gilitschenski},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026},
url={https://openreview.net/forum?id=4vmm8mlHkS}
}