2719. PyVRP: An improved high-performance VRP solver
Invited abstract in session WB-56: Heuristics for Vehicle Routing 2, stream Vehicle Routing and Logistics.
Wednesday, 10:30-12:00Room: Liberty 1.11
Authors (first author is the speaker)
| 1. | Leon Lan
|
| Vrije Universiteit Amsterdam | |
| 2. | Niels Wouda
|
| Technology and Operations Management, Rotterdam School of Management | |
| 3. | Wouter Kool
|
| ORTEC |
Abstract
This abstract presents an improved version of PyVRP, an open-source, vehicle routing problem (VRP) solver. PyVRP is designed to be user-friendly yet performant: performance-critical parts of the algorithm are implemented in C++, whereas all other parts are implemented in Python. The initial release of PyVRP in 2023 supported only the basic capacitated VRP (CVRP) and VRP with time windows (VRPTW). Since then, PyVRP has been extended to support different objectives, heterogeneous fleets, multiple depots, optional clients, simultaneous linehauls and backhauls, routing profiles, client groups and multiple trips. These features can be used to model many problem variants from the literature, as well as their combinations. Inspired by the recent successes of iterated local search (ILS) in solving VRPs, PyVRP has also moved from a hybrid genetic search (HGS) to an ILS algorithm. This shift has significantly improved PyVRP’s performance, particularly on large-scale instances with thousands of nodes, while also being conceptually much simpler than HGS. Benchmarks show that PyVRP achieves solutions within 1% of reported results across a wide variety of problem variants. With ongoing development and improvements, PyVRP continues to provide an accessible and high-performance VRP solver for researchers and practitioners in the field.
Keywords
- Vehicle Routing
- Software
Status: accepted
Back to the list of papers