Deep Implicit Coordination Graphs for Multi-agent Reinforcement Learning

Multi-agent reinforcement learning (MARL) requires coordination to efficiently solve certain tasks. Fully centralized control is often infeasible in such domains due to the size of joint action spaces. Coordination graph formalizations allow reasoning about the joint action based on the structure of interactions. However, they often require domain expertise in their design and can be difficult for dynamic environments with changing coordination requirements. This paper introduces the deep implicit coordination graph (DICG) architecture for such scenarios. DICG consists of a module for inferring the dynamic coordination graph structure which is then used by a graph neural network module to learn to implicitly reason about the joint actions or values. DICG allows learning the tradeoff between full centralization and decentralization via standard actor-critic methods to significantly improve coordination for domains with large number of agents. We apply DICG to both centralized-training-centralized-execution and centralized-trainingdecentralized-execution regimes. We demonstrate that DICG solves the relative overgeneralization pathology in predatory-prey tasks as well as outperforms various MARL baselines on the challenging StarCraft II Multi-agent Challenge (SMAC) and traffic junction environments.

Authors

• Sheng Li
• Jayesh K. Gupta
• Peter Morales Microsoft
• Ross Allen Massachusetts Institute of Technology
• Mykel J. Kochenderfer Stanford University

Keywords

• Multi-agent System
• Coordination
• Graph Neural Network
• Deep
• Reinforcement Learning