Tour length estimation guided vehicle routing
Loading...
Authors
Feldman, Stephen, I
Issue Date
2025-12
Type
Electronic thesis
Thesis
Thesis
Language
en_US
Keywords
Industrial and management engineering
Alternative Title
Abstract
The Vehicle Routing Problem (VRP) is a fundamental challenge in logistics, with most state-of-the-art heuristics relying on operators that directly manipulate explicit route sequences. This work proposes a heuristic method based on the 2-Stage Assignment-Routing formulation, which reframes the VRP as the problem of first partitioning customers among vehicles and then solving for the routes. The primary challenge of this formulation is the intractability of evaluating an assignment's cost, which requires solving a Traveling Salesman Problem (TSP). We address this challenge by replacing the exact TSP cost with a fast tour length estimator, creating a tractable objective function to guide a metaheuristic search. We evaluate a variety of existing estimators and introduce a novel Geometrically Assisted Regression Tree (GART), which employs machine learning to predict the TSP-to-MST cost ratio using a rich set of spatial and topological features. A comprehensive benchmark demonstrates that GART is significantly more precise than existing estimator models across a wide range of TSP instances. When integrated into a powerful VRP metaheuristic, the estimator-driven search improves upon initial solutions generated by constructive heuristics. However, a local optimizer of the estimator is not always consistent with the true VRP objective function. Our numerical experiments demonstrate that high-quality VRP solutions can be obtained by guiding the second stage with TSP tour length estimations rather than solving the full TSP directly. The effectiveness of this approach, however, depends critically on the precision of the estimation method: while reliable estimators enable strong performance, medium- to large-scale errors can cause the search to diverge from the true objective.
Description
December2025
School of Engineering
School of Engineering
Full Citation
Publisher
Rensselaer Polytechnic Institute, Troy, NY