Here is the code for our paper "Multihead Residual Value Factorization for Cooperative Multi-Agent Reinforcement Learning". You can run the experiments by the command below:

python src/main.py --config=mrmix --env-config=sc2 with env_args.map_name=3m

For the dependencies and SMAC benchmark, you can refer to pymarl2.

We use "res_qmix" for the name of our method at early stage. You can also run the experiments by

python src/main.py --config=res_qmix_*** --env-config=sc2 with env_args.map_name=3m

. It works identically as the former command.

We havn't update the name in some codes and filename for the reason of time, which will be updated later.