Rethinking the Implementation Tricks and Monotonicity Constraint in Cooperative Multi-agent Reinforcement Learning

Jian Hu, Siying Wang, Siyang Jiang, Weixun Wang

Keywords: multi-agent, reinforcement learning, experimental techniques, monotonicity

Abstract: QMIX, a very classical multi-agent reinforcement learning (MARL) algorithm, is often considered to be a weak performance baseline due to its representation capability limitations. However, we found that by improving the implementation techniques of QMIX we can enable it to achieve state-of-the-art under the StarCraft Multi-Agent Challenge (SMAC). Further, we found that the monotonicity constraint of QMIX is a key factor for its superior performance. We have open-sourced the code at https://github.com/xxxx/xxxx (Anonymous) for researchers to evaluate the effects of these proposed techniques. Our work has been widely used as a new QMIX baseline.

Blogpost Url: https://iclr-blogposts.github.io/2023/blog/2023/riit/

ICLR Papers: https://openreview.net/forum?id=X2MOcY_kIv

ID Of The Authors Of The ICLR Paper: ~Tabish_Rashid1

Conflict Of Interest: No

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2102.03479/code)