強化学習を用いた共同注視点に基づく合意形成

概要

論文の詳細を見る
"Focal Point" is a method of sharing information without explicit communication between cooperative multi-agents. It is hard for an artificial agent to prescribe the fixed algorithm which decides a focal point without relation 1D the environment. In this paper, we suppose the focal points as a distribution of a situation to induce the behavior of the agent. The autonomous agent can get focal points as a learning of the inside model of itself by the feedback from the environment. Consequently, the mutual agreement of multi-agent is emerged from an unity of the probabilistic choice of fields. Each agent is reinforced the filed potential value respectively by successful task. Finally, focal points can be acquired as potential distribution map of reinforcement value.
琉球大学工学部の論文