[2H4-GS-3c] 知識の利用：社会システム

[2H4-GS-3c-02] Experience Based Available Actions Network Generation

〇Yoshihiko Kato¹, Naoto Yoshida¹, Shuzo Koyama¹, Satoshi Kurihara² (1. Graduate School of Science and Technology, Keio University, 2. Faculty of Science and Technology, Keio University)

Keywords:Autonomous Agent, Knowledge Network

In order to realize an autonomous agent with the ability to select actions comparable to a human in a real environment, it is necessary to use an architecture with scalable data structures. Planning methods have been proposed that maintain a graph structure of symbolic actions and environmental states, and perform action selection so that we achieve such ability. However, these methods require defined symbolic actions, environments, and their relationships in advance, and it is impossible to do so in a real environment where there are countless kinds of concepts. In this study, we proposed an architecture that generates action logs, which are the experiences of agents, and builds relationships between actions and environmental states by connecting the action logs into a graph structure, and conducted experiments to see if the planning works. As a result, it was found that the planning functioned only when the bias of the behavior logs was small.