DT2GS

[NeurIPS 2023] -- Decompose a Task into Generalizable Subtasks in Multi-Agent Reinforcement Learning (DT2GS)

1.Installation

# install on-policy package
cd DT2GS
pip install -e .

2.Train/Transfer

run mappo

cd onpolicy/scripts/train_mappo_sh
chmod +x ./train_smac_3s5z_vs_3s6z.sh
./train_smac_3s5z_vs_3s6z.sh

run DT2GS

# run DT2GS or DT2GS_transfer
cd onpolicy/scripts/train_DT2GS_sh
chmod +x ./train_smac_subtask.sh
./train_smac_subtask.sh

# or
chmod +x ./transfer_smac_subtask.sh
./transfer_smac_subtask.sh

run UPDeT

# run UPDeT or UPDeT_transfer
cd onpolicy/scripts/train_DT2GS_sh
chmod +x ./train_smac_entity.sh
./train_smac_entity.sh

# or
chmod +x ./transfer_smac_entity.sh
./transfer_smac_entity.sh

run ASN_G or ASN_G_transfer

# run ASN_G or ASN_G_transfer
cd onpolicy/scripts/train_DT2GS_sh
chmod +x ./train_smac_asn_gatten.sh
./train_smac_asn_gatten.sh

# or
chmod +x ./transfer_smac_asn_gatten.sh
./transfer_smac_asn_gatten.sh

run ASN

cd onpolicy/scripts/train_DT2GS_sh
chmod +x ./train_smac_asn.sh
./train_smac_asn.sh

BibTeX

If you find our models useful, please consider citing our paper!

@article{tian2023decompose,
  title={Decompose a task into generalizable subtasks in multi-agent reinforcement learning},
  author={Tian, Zikang and Chen, Ruizhi and Hu, Xing and Li, Ling and Zhang, Rui and Wu, Fan and Peng, Shaohui and Guo, Jiaming and Du, Zidong and Guo, Qi and others},
  journal={Advances in Neural Information Processing Systems},
  volume={36},
  pages={78514--78532},
  year={2023}
}