Contrasting Centralized and Decentralized Critics in Multi-Agent Reinforcement Learning

TL;DR

This paper analyzes the differences between centralized and decentralized critics in multi-agent reinforcement learning, revealing misconceptions and providing practical insights through theoretical and empirical comparisons.

Contribution

It offers a formal analysis of critic types in multi-agent RL and empirically evaluates their advantages and disadvantages, challenging existing assumptions.

Findings

Centralized critics are not always superior to decentralized critics.

Both critic types have unique advantages and disadvantages.

Empirical results validate the theoretical analysis.

Abstract

Centralized Training for Decentralized Execution, where agents are trained offline using centralized information but execute in a decentralized manner online, has gained popularity in the multi-agent reinforcement learning community. In particular, actor-critic methods with a centralized critic and decentralized actors are a common instance of this idea. However, the implications of using a centralized critic in this context are not fully discussed and understood even though it is the standard choice of many algorithms. We therefore formally analyze centralized and decentralized critic approaches, providing a deeper understanding of the implications of critic choice. Because our theory makes unrealistic assumptions, we also empirically compare the centralized and decentralized critic methods over a wide set of environments to validate our theories and to provide practical advice. We show that there exist misconceptions regarding centralized critics in the current literature and show that the centralized critic design is not strictly beneficial, but rather both centralized and decentralized critics have different pros and cons that should be taken into account by algorithm designers.