Contrastive Learning-Based Agent Modeling for Deep Reinforcement Learning

Abstract

Multi-agent systems often require agents to collaborate with or compete against other agents with diverse goals, behaviors, or strategies. Agent modeling is essential when designing adaptive policies for intelligent machine agents in multi-agent systems, as this is the means by which the controlled agent (ego agent) understands other agents' (modeled agents) behavior and extracts their meaningful policy representations. These representations can be used to enhance the ego agent's adaptive policy which is trained by reinforcement learning. However, existing agent modeling approaches typically assume the availability of local observations from modeled agents during training or a long observation trajectory for policy adaption. To remove these constrictive assumptions and improve agent modeling performance, we devised a Contrastive Learning-based Agent Modeling (CLAM) method that relies only on the local observations from the ego agent during training and execution. With these observations, CLAM is capable of generating consistent high-quality policy representations in real time right from the beginning of each episode. We evaluated the efficacy of our approach in both cooperative and competitive multi-agent environments. The experiment results demonstrate that our approach improves reinforcement learning performance by at least 28% on cooperative and competitive tasks, which exceeds the state-of-the-art.