Curriculum Learning in Genetic Programming Guided Local Search for Large-scale Vehicle Routing Problems

TL;DR

This work tackles solving large-scale vehicle routing problems by learning GLS utility functions via Genetic Programming, augmented with Curriculum Learning to progressively expose the model to harder instances. The CL-GPGLS framework preserves the core GPGLS architecture while introducing a Curriculum Scheduler that sequences training data by difficulty and transitions between levels during training. Empirical results show CL-GPGLS outperforms random and fixed-scale strategies, with statistically significant improvements in final solutions and stability across 30 trials. The findings suggest curriculum-driven learning enhances generalization to large-scale VRPs and motivates future exploration of adaptive curricula and real-world datasets.

Abstract

Manually designing (meta-)heuristics for the Vehicle Routing Problem (VRP) is a challenging task that requires significant domain expertise. Recently, data-driven approaches have emerged as a promising solution, automatically learning heuristics that perform well on training instances and generalize to unseen test cases. Such an approach learns (meta-)heuristics that can perform well on the training instances, expecting it to generalize well on the unseen test instances. A recent method, named GPGLS, uses Genetic Programming (GP) to learn the utility function in Guided Local Search (GLS) and solved large scale VRP effectively. However, the selection of appropriate training instances during the learning process remains an open question, with most existing studies including GPGLS relying on random instance selection. To address this, we propose a novel method, CL-GPGLS, which integrates Curriculum Learning (CL) into GPGLS. Our approach leverages a predefined curriculum to introduce training instances progressively, starting with simpler tasks and gradually increasing complexity, enabling the model to better adapt and optimize for large-scale VRP (LSVRP). Extensive experiments verify the effectiveness of CL-GPGLS, demonstrating significant performance improvements over three baseline methods.

Figures (4)

• Figure 1: The evolutionary framework of CLGPGLS. The colored steps are the newly proposed steps compared with the original GPGLS.
• Figure 2: An example of the training process using the method proposed in this paper. The four steps correspond to four levels of difficulty in the training set.
• Figure 3: Test the fitness of four methods: ST (stochastic sequence), LO (large-only sequence), RG (random grouped sequence), and CL (curriculum learning sequence). The trend line shows that the method proposed in this paper achieves higher learning efficiency and outperforms the other three methods in the final generation.
• Figure 4: The box plots of the test performance of 30 independent runs of the proposed four methods.