The Surprising Effectiveness of PPO in Cooperative, Multi-Agent Games

TL;DR

This paper demonstrates that Proximal Policy Optimization (PPO), a simple on-policy reinforcement learning algorithm, performs surprisingly well in cooperative multi-agent environments, often rivaling or surpassing off-policy methods with minimal tuning.

Contribution

The study reveals that PPO can be an effective baseline for cooperative multi-agent reinforcement learning, challenging the belief that it is less sample efficient than off-policy algorithms.

Findings

PPO achieves strong performance across four multi-agent testbeds.

PPO often matches or exceeds off-policy methods in sample efficiency.

Implementation and hyperparameter choices are crucial for PPO's success.

Abstract

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}.