Automating the loop in traffic incident management on highway

TL;DR

The paper tackles the challenge of rapid, reliable decision-making in highway incident management. It introduces two AI-driven decision-support strategies: an LLM + Optimization hybrid that grounds decisions in a stochastic MDP and an IPS-based optimization, and a Full LLM that relies on Retrieval-Augmented Generation to reason with past incidents and a procedural manual. The LLM + Optimization approach demonstrates superior reliability in critical scenarios, while the Full LLM shows promising end-to-end automation but lacks robust guarantees. Overall, the work shows that integrating generative models with structured decision processes can yield practical, data-driven incident-management support with clear avenues for future improvement in reliability and explainability.

Abstract

Effective traffic incident management is essential for ensuring safety, minimizing congestion, and reducing response times in emergency situations. Traditional highway incident management relies heavily on radio room operators, who must make rapid, informed decisions in high-stakes environments. This paper proposes an innovative solution to support and enhance these decisions by integrating Large Language Models (LLMs) into a decision-support system for traffic incident management. We introduce two approaches: (1) an LLM + Optimization hybrid that leverages both the flexibility of natural language interaction and the robustness of optimization techniques, and (2) a Full LLM approach that autonomously generates decisions using only LLM capabilities. We tested our solutions using historical event data from Autostrade per l'Italia. Experimental results indicate that while both approaches show promise, the LLM + Optimization solution demonstrates superior reliability, making it particularly suited to critical applications where consistency and accuracy are paramount. This research highlights the potential for LLMs to transform highway incident management by enabling accessible, data-driven decision-making support.

Figures (6)

• Figure 1: LLM + Optimization Block pipeline.
• Figure 2: Graph representation of the system as a stochastic MDP, where nodes represent states and edges represent actions. Each edge is labeled with $n$, the number of times the action has been executed; $T$, the time associated with each action; and $P$, the probability of the action occurring.
• Figure 3: The update function for the Improved Prioritized Sweeping algorithm
• Figure 4: Full LLM pipeline.
• Figure 5: Box plot of evaluation scores for the two proposed solutions across three event types.