canyon / StarCraft

Implementations of QMIX and VDN on SMAC，corresponding to paper 《QMIX:Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning》 and 《Value-Decomposition Networks For Cooperative Multi-Agent Learning》

StarCraft

This is a pytorch implementation of the multi-agent reinforcement learning algrithms, QMIX and VDN, both of which are the state of art MARL algrithms. We trained these algrithms on SMAC, which is the decentralised micromanagement scenario of StarCraft II.

Requirements

• python=3.6.5
• torch=1.2.0
• SMAC
• pysc2

Acknowledgement

• SMAC
• pymarl

Quick Start

$ python main.py --evaluate_epoch=100 --map=3m

Directly run the main.py, then the two algrithms will be respectively tested on map '3m' for 100 episodes, using the pretrained model.

Result

Although qmix and vdn are the state of art multi-agent algrithms, they are unstable sometimes, you need to independently run several times to get better performence.

1. Win Rate of QMIX in Two Independent Runs on '3m'

2. Win Rate of VDN in Two Independent Runs on '3m'