Skip to content
/ CONS Public

The official code base of Cautiously-Optimistic kNowledge Sharing (AAAI 2024)

License

Notifications You must be signed in to change notification settings

byw0919/CONS

Repository files navigation

Cautiously-Optimistic Knowledge Sharing for Cooperative Multi-Agent Reinforcement Learning

The official code base of Cautiously-Optimistic kNowledge Sharing (CONS) (AAAI 2024)

Installation

Dependicies

  • gym==0.18.0
  • matplotlib==3.5.2
  • numpy==1.19.4
  • ray==0.8.6
  • six==1.16.0
  • torch==1.12.0

Clone this repository and install:

pip install -r requirements.txt

Traning

Patient Gold Miner (PGM)

  • PGM-6ag: PGM-6ag-v0
python main.py --alg cons --env_name ma_gym:PGM-6ag-v0 --max_episodes 80000
  • PGM-3ag: PGM-3ag-v0
python main.py --alg cons --env_name ma_gym:PGM-3ag-v0 --max_episodes 100000

Find the Treasure (FT)

  • Find the Treasure: FindTreasure-v0
python main.py --alg cons --env_name ma_gym:FindTreasure-v0 --max_episodes 50000 --reuse_network True --individual_rewards False --c_ep 30000 --c_w 0.4

Cleanup

python main_cleanup.py --max_episodes 80000 --c_w 0.4

Evaluating

some trained models are in model

PGM-6ag

python main.py --alg cons --env_name ma_gym:PGM-6ag-v0 --pkl_dir model/PGM-6ag/cons/ --load_model --evaluate --render

PGM-3ag

python main.py --alg cons --env_name ma_gym:PGM-3ag-v0 --pkl_dir model/PGM-3ag/cons/ --load_model --evaluate --render

Find the Treasure

python main.py --alg cons --env_name ma_gym:FindTreasure-v0 --reuse_network True --individual_rewards False --pkl_dir model/FindTreasure/cons/ --load_model --evaluate --render

Cleanup

python main_cleanup.py --pkl_dir model/Cleanup/cons/ --load_model --evaluate --render

Citation

Please cite our AAAI paper if you use this repository in your publications:

@inproceedings{ba2024cautiously,
  title={Cautiously-Optimistic Knowledge Sharing for Cooperative Multi-Agent Reinforcement Learning},
  author={Ba, Yanwen and Liu, Xuan and Chen, Xinning and Wang, Hao and Xu, Yang and Li, Kenli and Zhang, Shigeng},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={38},
  number={16},
  pages={17299--17307},
  year={2024}
}

Aknowledgement

About

The official code base of Cautiously-Optimistic kNowledge Sharing (AAAI 2024)

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages