[2I4-GS-5c] エージェント：基礎

[2I4-GS-5c-03] Proposition of Agent behavior selection framework in dynamic and complex environments

〇Daiki Shimokawa¹, Naoto Yoshida², Satoshi Kurihara¹ (1. Faculty of Science and Technology, Keio University, 2. Graduate School of Science and Technology, Keio University)

Keywords:AI, Planning, evolutionary algorithm

Currently, there is a lot of research on application-specific artificial intelligence, but the research of artificial general intelligence (AGI) is still in beginning phase. To construct AGI, high degree of adaptability in dynamic and complex environments is necessary, and to make this, multi-agent planning has been proposed. However, the agent network structure and the parameters for cooperative behavior have been determined manually in most of conventional methods. So, in this paper, we propose a method to automatically adjust these parameters by using an evolutional approach. As a result, it is found that the differential evolution method achieves a high degree of adaptation.