Toward LLM-agent-based modeling of transportation systems: A conceptual framework

Tianming Liu, Jirong Yang, Yafeng Yin

Abstract: In transportation system planning, agent-based models (ABMs) and microsimulations have become pivotal tools for understanding system dynamics and supporting decision-making. However, existing ABMs remain limited in their behavioral representation, flexibility, and reliance on comprehensive input data. This paper argues that recent advancements in large language models (LLMs) present a promising new avenue for enhancing agent-based transportation modeling. We propose an LLM-agent-based framework in which LLM agents act as behaviorally rich proxies for human travelers. By leveraging LLMs’ capabilities in natural language understanding, contextual reasoning, and generalization, our framework aims to overcome key limitations of traditional ABMs and unlock new modeling possibilities. We design LLM agents with structured profiles, memory systems, perception, decision-making, and action modules to align with the principles of activity-based travel demand modeling. Through system design and literature synthesis, we outline the conceptual and practical advantages of this approach and support our vision with a small-scale proof of concept simulation. Lastly, we discuss the remaining challenges and propose hybrid modeling as a near-term integration strategy. By positioning LLM agents as a novel and promising paradigm, we aim to expand the role of LLMs in agent-based transportation modeling and pave the way for new approaches to travel demand modeling.