[1P3-GS-7] Agents: Communications and agents

[1P3-GS-7-04] Deep Reinforcement Learning Based on Opponent Model for Automated Negotiating Agents

〇Toki Takahashi¹, Katsuhide Fujita¹ (1. Tokyo University of Agriculture and Technology)

Keywords:Multi-Agent Systems, Automated Negotiations, Reinforcement Learning

Research on automated negotiation, which is a means of resolving conflicts and forming consensus between autonomous agents, has been actively conducted, and various strategies of negotiating agents have been proposed.In this paper, we propose a deep reinforcement learning method to develop a negotiating agent that can respond to various negotiation strategies and negotiation scenarios.In addition, we propose an environment for deep reinforcement learning that uses information obtained by opponent model, which is a method of estimating the utility function of an opponent, as a state.As a result of the negotiation simulation experiment, it was confirmed that the agent who learned by the proposed method gained significantly higher individual utility than the agent who learned without opponent model.