Learning Multi-Agent Communication with Contrastive Learning
This work addresses the problem of coordination in multi-agent systems for researchers and practitioners, offering an incremental improvement by applying contrastive learning to communication learning.
The paper tackles the challenge of inducing effective communication in decentralized multi-agent reinforcement learning by treating messages as incomplete views of the environment and using contrastive learning to maximize mutual information between them. The method outperforms previous work in performance and learning speed in communication-essential environments, as shown through qualitative metrics and representation probing.
Communication is a powerful tool for coordination in multi-agent RL. But inducing an effective, common language is a difficult challenge, particularly in the decentralized setting. In this work, we introduce an alternative perspective where communicative messages sent between agents are considered as different incomplete views of the environment state. By examining the relationship between messages sent and received, we propose to learn to communicate using contrastive learning to maximize the mutual information between messages of a given trajectory. In communication-essential environments, our method outperforms previous work in both performance and learning speed. Using qualitative metrics and representation probing, we show that our method induces more symmetric communication and captures global state information from the environment. Overall, we show the power of contrastive learning and the importance of leveraging messages as encodings for effective communication.