Learning Individual Intrinsic Reward in MARL

This repository is an implementation of LIIR: Learning Individual Intrinsic Reward in Multi-Agent Reinforcement Learning. The framework for LIIR is inherited from PyMARL. LIIR is written in PyTorch and uses SMAC as its environment.

@inproceedings{
du2019learning,
title={LIIR: Learning Individual Intrinsic Reward in Multi-Agent Reinforcement Learning.},
author={Yali Du and Lei Han and Meng Fang and Tianhong Dai and Ji Liu and Dacheng Tao},
booktitle={Advances in Neural Information Processing Systems},
year={2019},
}

Setup

Set up the working environment:

pip install -r requirements.txt 

Set up the StarCraftII game core

bash install_sc2.sh  

Run an experiment

To train LIIR on the map with 3 marine,

python3 src/main.py --config=liir_smac --env-config=sc2 --map=3m  

All results will be stored in the Results folder.

Licence

The MIT License