Multi-agent reinforcement learning (MARL) has witnessed significant progress with the development of value function factorization methods. It allows optimizing a joint action-value function through the maximization of factorized per-agent utilities. In this paper, we show that in partially observable MARL problems, an agent's ordering over its own actions could impose concurrent constraints (across different states) on the representable function class, causing significant estimation errors during training. We tackle this limitation and propose PAC, a new framework leveraging Assistive information generated from Counterfactual Predictions of optimal joint action selection, which enable explicit assistance to value function factorization through a novel counterfactual loss. A variational inference-based information encoding method is developed to collect and encode the counterfactual predictions from an estimated baseline. To enable decentralized execution, we also derive factorized per-agent policies inspired by a maximum-entropy MARL framework. We evaluate the proposed PAC on multi-agent predator-prey and a set of StarCraft II micromanagement tasks. Empirical results demonstrate improved results of PAC over state-of-the-art value-based and policy-based multi-agent reinforcement learning algorithms on all benchmarks.

Tackling overestimation in Q-learning is an important problem that has been extensively studied in single-agent reinforcement learning, but has received comparatively little attention in the multi-agent setting. In this work, we empirically demonstrate that QMIX, a popular Q-learning algorithm for cooperative multiagent reinforcement learning (MARL), suffers from a more severe overestimation in practice than previously acknowledged, and is not mitigated by existing approaches. We rectify this with a novel regularization-based update scheme that penalizes large joint action-values that deviate from a baseline and demonstrate its effectiveness in stabilizing learning. Furthermore, we propose to employ a softmax operator, which we efficiently approximate in a novel way in the multiagent setting, to further reduce the potential overestimation bias. Our approach, Regularized Softmax (RES) Deep Multi-Agent Q-Learning, is general and can be applied to any Q-learning based MARL algorithm. We demonstrate that, when applied to QMIX, RES avoids severe overestimation and significantly improves performance, yielding state-of-the-art results in a variety of cooperative multi-agent tasks, including the challenging StarCraft II micromanagement benchmarks.

Cooperative multi-agent reinforcement learning (MARL) has been proposed for multi-agent collaborations toaccomplish manychallenging tasks[1,2,3,4].

Table 1: Payoff matrix of the one-step multi-state non-monotonic cooperative matrix game and reconstructed resultsfromcorresponding baselines.

To enable decentralized execution, we alsoderivefactorized per-agentpolicies inspired byamaximum-entropyMARL framework.

In this paradigm of centralised training for decentralised execution, QMIX [25] is a popular Qlearning algorithm with state-of-the-art performance ontheStarCraft Multi-Agent Challenge [26]. QMIX represents the optimal joint action value function using a monotonicmixing function of per-agent utilities.

Efficient exploration in deep cooperativemulti-agent reinforcement learning (MARL) still remains challenging in complex coordination problems.

Efficient exploration in deep cooperativemulti-agent reinforcement learning (MARL) still remains challenging in complex coordination problems.