SOG

Code for NeurIPS2022 accepted paper: Self-Organized Group for Cooperative Multi-agent Reinforcement Learning.

This codebase is built on top of the PyMARL framework and the codebase of REFIL algorithm. Thanks for Shariq Iqbal for sharing his code.

Setup instructions

Please follow the instructions in REFIL codebase. Note: If you want to run environment sc2custom, an empty map needs to be copied to the SC2 directory. Note: The particle environment uses gym==0.10.5.

Run an experiment

Run an ALGORITHM from the folder src/config/algs in an ENVIRONMENT from the folder src/config/envs

export CUDA_VISIBLE_DEVICES="0" && python src/main.py --env-config=<ENVIRONMENT> --config=<ALGORITHM> with <PARAMETERS>

Possible ENVIRONMENTs are:

• particle: Resource collection and Predator-prey environment from the paper. The default is Resource collection. Add option env_args.scenario_id="predator_prey.py" for using the Predator-prey environment.
• sc2custom: StarCraft environment from the paper

Possible ALGORITHMs are:

• rlsocom: SOG-rl in the paper (recommended);
• dppsocom: SOG-dpp in the paper;
• socom: SOG in the paper;
• copa: COPA in the paper;
• qmix_atten: A-QMIX in the paper;
• qmix_atten_gat: MAGIC in Appendix I;
• qmix_atten_silgat: Gated-ACML in Appendix I.

Command examples

Run SOG-rl with environment Resource collection:

export CUDA_VISIBLE_DEVICES="0" && python src/main.py --env-config=particle --config=rlsocom with train_map_num=[2,3,4,5] test_map_num=[6,7,8]

If you want to use this repository, please consider citing:

@inproceedings{shaoself,
  title={Self-Organized Group for Cooperative Multi-agent Reinforcement Learning},
  author={Shao, Jianzhun and Lou, Zhiqiang and Zhang, Hongchang and Jiang, Yuhang and He, Shuncheng and Ji, Xiangyang},
  booktitle={Advances in Neural Information Processing Systems}
}