Counterfactual Multi-Agent Policy Gradients

TL;DR

This paper introduces COMA, a multi-agent reinforcement learning method with a counterfactual baseline that improves decentralized policy learning in complex environments like StarCraft.

Contribution

The paper proposes COMA, a novel multi-agent actor-critic algorithm using a counterfactual baseline for efficient credit assignment and improved decentralized policy optimization.

Findings

COMA outperforms other multi-agent actor-critic methods in StarCraft micromanagement tasks.

COMA's performance is competitive with centralized controllers with full state access.

The counterfactual baseline effectively addresses multi-agent credit assignment challenges.

Abstract

Cooperative multi-agent systems can be naturally used to model many real world problems, such as network packet routing and the coordination of autonomous vehicles. There is a great need for new reinforcement learning methods that can efficiently learn decentralised policies for such systems. To this end, we propose a new multi-agent actor-critic method called counterfactual multi-agent (COMA) policy gradients. COMA uses a centralised critic to estimate the Q-function and decentralised actors to optimise the agents' policies. In addition, to address the challenges of multi-agent credit assignment, it uses a counterfactual baseline that marginalises out a single agent's action, while keeping the other agents' actions fixed. COMA also uses a critic representation that allows the counterfactual baseline to be computed efficiently in a single forward pass. We evaluate COMA in the testbed of StarCraft unit micromanagement, using a decentralised variant with significant partial observability. COMA significantly improves average performance over other multi-agent actor-critic methods in this setting, and the best performing agents are competitive with state-of-the-art centralised controllers that get access to the full state.