Musteata, Denisa-Gabriela ("Gheorghe Asachi” Technical University of Iasi), Porcarasu, Silvia („Gheorghe Asachi” TECHNICAL UNIVERSITY of IASI), Lazar, Razvan-Gabriel (Gheorghe Asachi Technical University of Iasi), Caruntu, Constantin-Florin (Gheorghe Asachi Technical University of Iasi)

LLM-Based Control Policies for Intelligent Traffic Intersections in SUMO-Veins

Scheduled for presentation during the Regular session "Automotive and Traffic" (ThA1), Thursday, October 22, 2026, 12:15−12:30, Magnum 1/2

2026 30th International Conference on System Theory, Control and Computing (ICSTCC), October 21-24, 2026, Iasi, Romania

This information is tentative and subject to change. Compiled on October 13, 2026

Abstract

Large Language Models (LLMs) have demonstrated strong potential for supporting decision-making across many scenarios. In particular, the increasing complexity and dynamic nature of Intelligent Transportation Systems (ITSs) create a need for advanced decision-support mechanisms capable of adapting to rapidly changing traffic conditions. In this context, LLMs offer a promising approach for processing heterogeneous information and generating adaptive control strategies. This paper presents a modular architecture that integrates an LLM-based control system into a traffic simulation environment by connecting Simulation of Urban Mobility (SUMO) and OMNeT++/Veins through an HTTP interface. The proposed approach enables traffic state exchange between an external LLM and simulated vehicles. The LLM acts as a vehicle-level decision-making controller that generates longitudinal driving actions (accelerate, brake, and keep speed) for individual vehicles based on their local traffic state, rather than controlling traffic signal timing. The results indicate that LLM-based control policies can adapt to varying traffic conditions while maintaining stable traffic flow in the SUMO-Veins co-simulation environment.