Most recently developed approaches to cooperative multi-agent reinforcement learning in the \emph{centralized training with decentralized execution} setting involve estimating a centralized, joint value function. In this paper, we demonstrate that, despite its various theoretical shortcomings, Independent PPO (IPPO), a form of independent learning in which each agent simply estimates its local value function, can perform just as well as or better than state-of-the-art joint learning approaches on popular multi-agent benchmark suite SMAC with little hyperparameter tuning. We also compare IPPO to several variants; the results suggest that IPPO's strong performance may be due to its robustness to some forms of environment non-stationarity.
@article{arxiv.2011.09533,
title = {Is Independent Learning All You Need in the StarCraft Multi-Agent Challenge?},
author = {Christian Schroeder de Witt and Tarun Gupta and Denys Makoviichuk and Viktor Makoviychuk and Philip H. S. Torr and Mingfei Sun and Shimon Whiteson},
journal= {arXiv preprint arXiv:2011.09533},
year = {2020}
}