Locke637 / CDS

[NeurIPS 2021] CDS achieves remarkable success in challenging benchmarks SMAC and GRF by balancing sharing and diversity.

CDS: Celebrating Diversity in Shared Multi-Agent Reinforcement Learning

The paper is now available in arXiv and accepted by NeurIPS 2021. Our approach can help both value-based and policy-based baselines (such as QMIX, QPLEX, and MAPPO) to explore sophisticated strategies for improving learning efficiency in challenging benchmarks.

Note

This codebase accompanies the paper submission "Celebrating Diversity in Shared Multi-Agent Reinforcement Learning"(CDS website) and is based on GRF, PyMARL and SMAC codebases which are open-sourced.

Publication

If you find this repository useful, please cite our paper:

@article{chenghao2021celebrating,
  title={Celebrating diversity in shared multi-agent reinforcement learning},
  author={Chenghao, Li and Wang, Tonghan and Wu, Chengjie and Zhao, Qianchuan and Yang, Jun and Zhang, Chongjie},
  journal={Advances in Neural Information Processing Systems},
  volume={34},
  year={2021}
}