Interactive self-reflection based reinforcement learning for multiagent coordination

摘要

One of the main goals of Artificial Intelligence is to realizing an intelligent agent that behaves autonomously by its sense of value. Reinforcement learning (RL, in short) is the major learning mechanism for the agent to adapt itself to various situations of an unknown environment flexibly. The merit of RL is that the design cost is small. Only giving the agent a goal state as setting a reward, a set of optimal behavior sequences toward the goal state from each state can be learned by trial and error. However, in a multiagent system (MAS, in short) environment that has mutual dependency among agents, it is difficult for a human to setup suitable learning goals for each agent, besides the existent framework of RL that aims for objective and egoistic optimality is inadequate. Therefore, it requires the active and interactive learning function that treats how to coordinate the interaction among other learning agents. This paper presents a new framework of reinforcement learning to generate and coordinate the learning goals interactively among agents. Then coordinating conflict of interests among the agents is discussed to dissolve the social dilemma.