QMIX: Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning

TLDR

In many real‑world settings agents must coordinate while acting decentralised, yet can be trained centrally with global state; learning joint action‑values with extra state is attractive but extracting decentralised policies is unclear. QMIX is a novel value‑based method that trains decentralised policies in a centralised end‑to‑end fashion. QMIX estimates joint action‑values as a monotonic non‑linear combination of per‑agent local‑observation‑based values, enabling tractable maximisation and consistency between centralised and decentralised policies. On StarCraft II micromanagement tasks, QMIX significantly outperforms existing value‑based multi‑agent reinforcement learning methods.

Abstract

In many real-world settings, a team of agents must coordinate their behaviour while acting in a decentralised way. At the same time, it is often possible to train the agents in a centralised fashion in a simulated or laboratory setting, where global state information is available and communication constraints are lifted. Learning joint action-values conditioned on extra state information is an attractive way to exploit centralised learning, but the best strategy for then extracting decentralised policies is unclear. Our solution is QMIX, a novel value-based method that can train decentralised policies in a centralised end-to-end fashion. QMIX employs a network that estimates joint action-values as a complex non-linear combination of per-agent values that condition only on local observations. We structurally enforce that the joint-action value is monotonic in the per-agent values, which allows tractable maximisation of the joint action-value in off-policy learning, and guarantees consistency between the centralised and decentralised policies. We evaluate QMIX on a challenging set of StarCraft II micromanagement tasks, and show that QMIX significantly outperforms existing value-based multi-agent reinforcement learning methods.