The Surprising Effectiveness of PPO in Cooperative, Multi-Agent Games
This provides a strong, simple baseline for researchers and practitioners in cooperative multi-agent reinforcement learning, though it is incremental as it re-evaluates an existing method.
The paper tackles the problem of PPO's perceived inefficiency in multi-agent reinforcement learning by showing that with proper tuning, PPO-based methods achieve competitive or superior performance in cooperative multi-agent games, matching or outperforming off-policy methods in final returns and sample efficiency across four testbeds.
Proximal Policy Optimization (PPO) is a ubiquitous on-policy reinforcement learning algorithm but is significantly less utilized than off-policy learning algorithms in multi-agent settings. This is often due to the belief that PPO is significantly less sample efficient than off-policy methods in multi-agent systems. In this work, we carefully study the performance of PPO in cooperative multi-agent settings. We show that PPO-based multi-agent algorithms achieve surprisingly strong performance in four popular multi-agent testbeds: the particle-world environments, the StarCraft multi-agent challenge, Google Research Football, and the Hanabi challenge, with minimal hyperparameter tuning and without any domain-specific algorithmic modifications or architectures. Importantly, compared to competitive off-policy methods, PPO often achieves competitive or superior results in both final returns and sample efficiency. Finally, through ablation studies, we analyze implementation and hyperparameter factors that are critical to PPO's empirical performance, and give concrete practical suggestions regarding these factors. Our results show that when using these practices, simple PPO-based methods can be a strong baseline in cooperative multi-agent reinforcement learning. Source code is released at \url{https://github.com/marlbenchmark/on-policy}.