The implementation of the following methods can be found in this codebase:
-
- install SMAC following https://github.com/oxwhirl/smac
-
- install Multi-Agent MuJoCo following https://github.com/schroederdewitt/multiagent_mujoco
-
- install MPE following https://github.com/openai/multiagent-particle-envs
-
- install required packages: pip install -r requirements.txt
python3 on-policy-main/train_smac.py --map_name 2s3z --use_eval --penalty_method True --dtar_kl 0.02 --experiment_name dtar_0.02_V_penalty_2M --num_env_steps 2000000 --group_name dpo --seed 1 --multi_rollout True --n_rollout_threads 1
Here, we provide results in three different SMAC scenarios using default hyperparameters.
--->
If you are using the codes, please cite our papers.
@article{DPO,
title={A Fully Decentralized Surrogate for Multi-Agent Policy Optimization},
author={Su, Kefan and Lu, Zongqing},
journal={Transactions on Machine Learning Research},
year={2024}
}