Title: Leveraging Large Language Models for Transportation Systems Management and Operations: Bridging Textual Reasoning and Multi-modal Decision Support

Abstract:

The management and operations (TSMO) of transportation systems are becoming increasingly reliant on the prompt analysis of diverse data sources. These inputs range from continuous sensor feeds and visual records to incident documentation and traveler feedback. Large language models (LLMs), particularly the emerging class of multi-modal large language models (MM-LLMs), are introducing novel capabilities for synthesizing both structured and unstructured information into actionable decision support tools for operators. This review examines applications of LLMs and MM-LLMs within TSMO, categorizing them into three primary areas: transportation operations and services (supply side), mobility and fleet services (demand side), and data analytics, modeling, and decision support.

Employing a screening process guided by PRISMA standards, this study consolidates current research findings while clearly separating practical, operationally focused applications from those that remain at the prototype stage or are conceptual. The paper highlights persistent obstacles, including data heterogeneity, the need for real-time inference, model explainability, the fusion of multi-modal data, and governance issues. Furthermore, it delineates current limitations and proposes future research avenues such as localized adaptation, edge computing deployment, standardized benchmarking, and enhanced collaboration among agencies. Ultimately, the analysis suggests that LLM-based technologies hold significant promise as a decision-support layer, with MM-LLMs providing distinct advantages in scenarios requiring the integration of varied text, visual, and sensor data streams.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC