Route Optimization Reimagined: Multi-Modal Large Language Models for Next-Generation Vehicle Routing

Abstract

The Capacitated Vehicle Routing Problem (CVRP) is a fundamental combinatorial optimization problem in logistics that requires the efficient routing of vehicles under capacity constraints. Traditional approaches, such as exact algorithms, heuristics, and Machine Learning (ML) methods, have achieved notable success, but often struggle with scalability, adaptability, and integration of heterogeneous data in dynamic environments. Large Language Models (LLMs) have recently demonstrated potential for solving such optimization tasks by generating heuristics and enabling structured reasoning through natural language. However, their reliance on purely textual inputs limits their ability to fully capture multidimensional information, such as spatial layouts and real-time sensor data. Multi-modal Large Language Models (MLLMs) extend LLMs by incorporating text, images, and numerical data, offering a more comprehensive framework for solving complex optimization problems, such as CVRP. Despite their potential, the application of MLLMs in CVRP remains underexplored. This paper reviews recent advancements at the intersection of MLLMs and routing optimization, evaluates their current effectiveness, and outlines open challenges and future research directions. These insights aim to support the development of scalable, interpretable, and adaptable solutions for real-world logistics and transportation systems.