This repository implements EXPODE used in the paper "EXPODE: EXploiting POlicy Discrepancy for Efficient Exploration in Multi-agent Reinforcement Learning" (https://dl.acm.org/doi/10.5555/3545946.3598619).

Set up StarCraft II and SMAC:

cd EXPODE-master/pymarl
bash install_sc2.sh

This will download SC2 into the 3rdparty folder and copy the maps necessary to run over. We use a modified version of SMAC, which can be installed by following commands:

cd EXPODE-master/QPLEX_smac_env
pip install -e .

The requirements.txt file can be used to install the necessary packages into a virtual python environment.

In this code, We evaluate our method on two environments: Predator and Prey("pred_prey_punish"), SMAC("sc2"). We use the default settings in SMAC, and the results in our paper use Version SC2.4.6.2.69232.

To train EXPODE on Predator and Prey, run the following command:

cd EXPODE-master/pymarl
python3 src/main.py --config=EXPODE_toygame --env-config=pred_prey_punish with env_args.map_name=origin 

To train EXPODE on SC2 setting tasks, run the following command:

python3 src/main.py --config=EXPODE_sc2 --env-config=sc2 with env_args.map_name=MMM2 

Map names for SMAC include 2s3z, 3s5z, 5m_vs_6m, 3s5z_vs_3s6z, MMM2, 3s_vs_5z, 6h_vs_8z, corridor, and 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

Our experimental results are saved in the EXPODE_experimental_data folder, and can be found by the following command:

cd EXPODE-master/pymarl/src/EXPODE_experimental_data

Then, the figures will be shown by running python script:

python test_plot_sc2.py
python test_plot_stag.py