This code is build based on PyMARL2 and PyMARL. We assume that you have experience with PyMARL. The requirements are the same as PyMARL2.

cd ResQ
python3 src/main.py --config=ResQ --env-config=sc2 with env_args.map_name=2s3z

cd ResZ
python3 src/main.py --config=ResZ --env-config=sc2 with env_args.map_name=2s3z

The config files act as defaults for an algorithm or environment.

They are all located in src/config. --config refers to the config files in src/config/algs --env-config refers to the config files in src/config/envs

All results will be stored in the Results folder.

If you use ResQ/ResZ in your research, please cite our paper

Siqi Shen, Mengwei Qiu, Jun Liu, Weiquan Liu, Yongquan Fu, Xinwang Liu, Cheng Wang. ResQ: A Residual Q Function-based Approach for Multi-Agent Reinforcement Learning Value Factorization. NeurIPS 2022

In BibTeX format:

@inproceedings{ResQ, 
author = {Siqi Shen and Mengwei Qiu and Jun Liu and Weiquan Liu and Yongquan Fu and Xinwang Liu and Cheng Wang}, 
title = {ResQ: A Residual Q Function-based Approach for Multi-Agent Reinforcement Learning Value Factorization}, 
booktitle = {{NeurIPS}}, 
year = {2022}}

Code licensed under the Apache License v2.0