Learning to Solve Vehicle Routing Problems: A Survey

Abstract

This paper provides a systematic overview of machine learning methods applied to solve NP-hard Vehicle Routing Problems (VRPs). Recently, there has been a great interest from both machine learning and operations research communities to solve VRPs either by pure learning methods or by combining them with the traditional hand-crafted heuristics. We present the taxonomy of the studies for learning paradigms, solution structures, underlying models, and algorithms. We present in detail the results of the state-of-the-art methods demonstrating their competitiveness with the traditional methods. The paper outlines the future research directions to incorporate learning-based solutions to overcome the challenges of modern transportation systems.

Neural combinatorial optimization: from routing to integer programming

“…To overcome the above issues, this thesis centers on developing NCO methods to solve COPs, i.e., from routing problems (which is a popular class of COPs) to integer programs (which can model general COPs). We first target at routing problems which are mostly studied in NCO literature [6,27]. We select routing problems rather other COPs because 1) routing problems are one class of COPs which are solved very often in daily life; 2) despite a long history of study, routing problems are still hard to solve optimally or efficiently, especially for large-scale ones, and deep learning may have chances to enhance the performance; 3) the research on routing problems has the potential to benefit the development of NCO methods for similar COPs, e.g., scheduling, bin packing [12,28,29].…”

Learning generalizable heuristics for solving vehicle routing problem under distribution shift

“…To address these issues, this thesis focuses on developing neural heuristics to solve routing problems under distribution shifts. Specifically, this thesis targets VRPs which are mostly studied in the literature of learning to optimize [16,21]. The research on VRPs is promising to boost the development of neural heuristics for similar combinatorial optimization problems, e.g., 3D bin packing and resource/task scheduling [22][23][24].…”