[2P1-OS-23] 社会科学・人文科学分野の行動インサイトを活用した機械学習と最適化

[2P1-OS-23-03] Pedestrian model introducing a diversity of potential preferences

~Toward the Generation of Explanations for mechanism of behavior change~

〇Keiju Tazawa¹, Sachiyo Arai¹ (1. Chiba University)

Keywords:Multi Agent Simulation, decision making, reinforcement learning

In recent years, data-driven urban planning that aggregates all kinds of urban data on a digital space, including 3D urban models, has been attracting attention. In this study, we focus on the design method of street space utilizing pedestrian flow data on these virtual urban spaces. Pedestrian flow data visualizes the flow of people in a street space, and is expected to be used to elucidate the mechanism of pedestrian behavior change. For this purpose, it is necessary to model pedestrians and to understand the changes in behavior due to differences in pedestrians' preferences for the street space. In existing studies, various pedestrian models have been developed using multi-agent simulation (MAS), but all of them have uniform behavioral norms for pedestrians and cannot represent the differences in preferences among people.
In this study, we propose a pedestrian model that can express the diversity of preferences, observe changes in behavior among individuals when the model is introduced, and verify the importance of expressing differences in preferences. The proposed model is formulated as a multi-objective sequential decision-making problem in which the process of determining a route based on various preferences that pedestrians have latent.