Deep Multi-Agent Reinforcement Learning with Relevance Graphs

Abstract

Over recent years, deep reinforcement learning has shown strong successes in complex single-agent tasks, and more recently this approach has also been applied to multi-agent domains. In this paper, we propose a novel approach, called MAGnet, to multi-agent reinforcement learning (MARL) that utilizes a relevance graph representation of the environment obtained by a self-attention mechanism, and a message-generation technique inspired by the NerveNet architecture. We applied our MAGnet approach to the Pommerman game and the results show that it significantly outperforms state-of-the-art MARL solutions, including DQN, MADDPG, and MCTS.

arXiv commentThe first two authors contributed equally. Author ordering determined by coin flip over a Google Hangout. Accepted at NIPS 2018 Deep RL Workshop

References in corpus (4)

• Continuous control with deep reinforcement learning
• Playing Atari with Deep Reinforcement Learning
• Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environments
• Learning to Search with MCTSnets

Cited by in corpus (9)

• QMIX: Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning
• Deep Implicit Coordination Graphs for Multi-agent Reinforcement Learning
• Graph Convolutional Reinforcement Learning
• Evolutionary Population Curriculum for Scaling Multi-Agent Reinforcement Learning
• Accelerating Training in Pommerman with Imitation and Reinforcement Learning
• Safer Deep RL with Shallow MCTS: A Case Study in Pommerman
• Learning to Optimize Industry-Scale Dynamic Pickup and Delivery Problems
• DAIR: Disentangled Attention Intrinsic Regularization for Safe and Efficient Bimanual Manipulation
• Developing a Successful Bomberman Agent