QPLEX: Duplex Dueling Multi-Agent Q-Learning

TLDR

Value‑based multi‑agent reinforcement learning often uses centralized training with decentralized execution, relying on the Individual‑Global‑Max principle to align joint and local actions, but existing scalable methods either restrict value function expressiveness or relax IGM consistency, risking instability or poor performance in complex domains. The authors propose QPLEX, a duplex dueling multi‑agent Q‑learning method that factorizes the joint value function using a duplex dueling network. QPLEX employs a duplex dueling network that embeds the IGM principle into its architecture, enabling efficient learning of the joint value function. Theoretical analysis confirms that QPLEX realizes a complete IGM function class, and experiments on StarCraft II micromanagement tasks show it outperforms state‑of‑the‑art baselines, achieves high sample efficiency, and benefits from offline data without extra online exploration.

Abstract

We explore value-based multi-agent reinforcement learning (MARL) in the popular paradigm of centralized training with decentralized execution (CTDE). CTDE has an important concept, Individual-Global-Max (IGM) principle, which requires the consistency between joint and local action selections to support efficient local decision-making. However, in order to achieve scalability, existing MARL methods either limit representation expressiveness of their value function classes or relax the IGM consistency, which may suffer from instability risk or may not perform well in complex domains. This paper presents a novel MARL approach, called duPLEX dueling multi-agent Q-learning (QPLEX), which takes a duplex dueling network architecture to factorize the joint value function. This duplex dueling structure encodes the IGM principle into the neural network architecture and thus enables efficient value function learning. Theoretical analysis shows that QPLEX achieves a complete IGM function class. Empirical experiments on StarCraft II micromanagement tasks demonstrate that QPLEX significantly outperforms state-of-the-art baselines in both online and offline data collection settings, and also reveal that QPLEX achieves high sample efficiency and can benefit from offline datasets without additional online exploration.