Cooperative Behavior Acquisition for Multi-agent Systems by Q-learning

In this paper, we focused on the problem of "trash collection", in which multiple agents collect all trash as quickly as possible. The goal of the present research is for multiple agents to learn to accomplish a task by interacting with the environment and acquiring cooperative behavior rules. We construct the learning agent using Q-learning, which is a representative technique of reinforcement learning. Q-learning is designed to find a policy that maximizes the learning for all states. The decision policy is represented by a function. The goal is for multiple agents to learn to accomplish a task by interacting with the environment and other agents. The action value function is shared among agents. The effectiveness of the learning is verified experimentally