[2J6-GS-2] Machine learning: Advancement reinforcement learning (2)

[2J6-GS-2-05] Multi-Agent Deep Reinforcement Learning with Multi-Branch Networks Considering Interaction

〇Tsuyoshi Goto¹, Tsubasa Hirakawa¹, Takayoshi Yamashita¹, Hironobu Fujiyoshi¹ (1. Chubu University)

Keywords:Reinforcement Learning, Multi Agent, Interaction

When multiple agents are in the same environment, collisions may happen with each other. Because the agents consider their own interests or have a negative effect on other agents. In situation that happens these deadlocks, agents should select the action considering other agents based on multi-agent reinforcement learning which train multiple agents simultaneously. In this paper, we propose the method that trains multiple agents in a network with multi-branch network for this problem. It is possible to train an interaction between agents. In experiment, we build the environment that happens the deadlock between self-driving cars and compare with independent network of each agent. Moreover, we show the behavior of the agent in a deadlock situation.