大規模言語モデル駆動の都市交通シミュレーション基盤

This project was selected under JST PRESTO (Precursory Research for Embryonic Science and Technology) in the research area "Co-Creation of Transformation Platform Technology for Human and Society by Integration of the Humanities and Sciences." It aims to build a simulation platform that leverages the advanced reasoning capabilities and world knowledge of Large Language Models (LLMs) to more deeply understand and reproduce human behavior in urban environments.

In conventional traffic simulations, human behavior has been modeled based on predefined rules or statistical distributions. This research introduces LLMs as models of human behavior, working to reproduce flexible decision-making that responds to context and situations. Furthermore, by integrating real-world data obtained from pedestrian flow sensing with behavioral intention inference by LLMs, we aim to build a framework for understanding the human intentions and purposes behind observed movement patterns.