Is Independent Learning All You Need in the StarCraft Multi-Agent Challenge?
This work challenges the common assumption that centralized value functions are necessary for cooperative multi-agent reinforcement learning, providing a simpler, competitive baseline for researchers in multi-agent systems.
This paper investigates Independent PPO (IPPO) in the StarCraft Multi-Agent Challenge (SMAC), finding that it performs as well as or better than state-of-the-art joint learning approaches. The strong performance of IPPO is attributed to its robustness to environmental non-stationarity.
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.