Visualization of Cooperative Behavior of Social Agents Based on Interactive Reinforcement Learning

Abstract

In the multi-agent reinforcement learning, multiple agents are supposed to learn cooperative behavior. One unsolved problem is how to obtain an appropriate cooperative behavior from the mutual interaction during the reinforcement learning process. In this research, we propose an interactive reinforcement learning system with the efficient cooperative ability through the social interaction among agents. It is effective to visualize the satate and agent behavior of learning process to analyze the interaction system. Therefore, we apply the visualizing tool, which can reflect the idea to the agent model immediately. The cooperative process changing dynamically is analyzed by indicating the mutual interaction among agents on the computer.