Solving the 3L-CVRP (as proposed by Gendreau et al.) with reinforcement learning (6-month MSc Thesis)

Citing

Paper (no appendix): https://www.sciencedirect.com/science/article/pii/S2405896325012571

Preprint with appendix: https://www.researchgate.net/publication/389511132_Towards_Scalable_Three-Dimensional_Loading_Capacitated_Vehicle_Routing

@article{SCHOEPF20252939,
title = {Towards Scalable Three-Dimensional Loading Capacitated Vehicle Routing},
journal = {IFAC-PapersOnLine},
volume = {59},
number = {10},
pages = {2939-2944},
year = {2025},
note = {11th IFAC Conference on Manufacturing Modelling, Management and Control MIM 2025},
issn = {2405-8963},
doi = {https://doi.org/10.1016/j.ifacol.2025.09.494},
url = {https://www.sciencedirect.com/science/article/pii/S2405896325012571},
author = {Stefan Schoepf and Stephen Mak and Julian Senoner and Liming Xu and Torbjörn Netland and Alexandra Brintrup},
keywords = {Supply Chain, Vehicle Routing, Optimisation, Reinforcement Learning, Logistics},
abstract = {Current vehicle routing methods suffer from non-linear scaling with increasing problem size and are therefore bound to limited geographic areas to compute results in time for day-to-day operations. This only allows for local optima in routing and leaves global optimization potential untouched. We develop a reinforcement learning model to approximately solve the three-dimensional loading capacitated vehicle routing problem in linear time. We demonstrate the linear time scaling of our reinforcement learning model and benchmark our routing performance against state-of-the-art methods. Our results show a promising first step towards large-scale logistics optimization with reinforcement learning.}
}