Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SVRPBench: A Realistic Benchmark for Stochastic Vehicle Routing Problem

Ahmed Heakl, Yahia Salaheldin Shaaban, Martin Takac, Salem Lahlou, Zangir Iklassov

TL;DR

SVRPBench tackles the gap between deterministic VRP benchmarks and real-world stochastic routing by introducing an urban-scale, open benchmark that embeds time-dependent congestion, log-normal delays, accident disruptions, and heterogeneous time windows. The authors develop a unified dataset construction pipeline with multi-depot/multi-vehicle support, diverse instance scales, and automated validation, accompanied by a standardized evaluation protocol and baseline solvers spanning classical, metaheuristic, industrial, and learning-based methods. Across 500+ instances, they show traditional solvers like OR-Tools achieve the best overall cost and feasibility, while RL approaches struggle with distributional shift and large-scale generalization, though multi-depot configurations generally improve performance for all methods. The work emphasizes realism and reproducibility by releasing data, code, and a public leaderboard, and it calls for adaptive, noise-aware routing algorithms that generalize beyond synthetic assumptions to enable robust, deployable logistics solutions.

Abstract

Robust routing under uncertainty is central to real-world logistics, yet most benchmarks assume static, idealized settings. We present SVRPBench, the first open benchmark to capture high-fidelity stochastic dynamics in vehicle routing at urban scale. Spanning more than 500 instances with up to 1000 customers, it simulates realistic delivery conditions: time-dependent congestion, log-normal delays, probabilistic accidents, and empirically grounded time windows for residential and commercial clients. Our pipeline generates diverse, constraint-rich scenarios, including multi-depot and multi-vehicle setups. Benchmarking reveals that state-of-the-art RL solvers like POMO and AM degrade by over 20% under distributional shift, while classical and metaheuristic methods remain robust. To enable reproducible research, we release the dataset and evaluation suite. SVRPBench challenges the community to design solvers that generalize beyond synthetic assumptions and adapt to real-world uncertainty.

SVRPBench: A Realistic Benchmark for Stochastic Vehicle Routing Problem

TL;DR

SVRPBench tackles the gap between deterministic VRP benchmarks and real-world stochastic routing by introducing an urban-scale, open benchmark that embeds time-dependent congestion, log-normal delays, accident disruptions, and heterogeneous time windows. The authors develop a unified dataset construction pipeline with multi-depot/multi-vehicle support, diverse instance scales, and automated validation, accompanied by a standardized evaluation protocol and baseline solvers spanning classical, metaheuristic, industrial, and learning-based methods. Across 500+ instances, they show traditional solvers like OR-Tools achieve the best overall cost and feasibility, while RL approaches struggle with distributional shift and large-scale generalization, though multi-depot configurations generally improve performance for all methods. The work emphasizes realism and reproducibility by releasing data, code, and a public leaderboard, and it calls for adaptive, noise-aware routing algorithms that generalize beyond synthetic assumptions to enable robust, deployable logistics solutions.

Abstract

Robust routing under uncertainty is central to real-world logistics, yet most benchmarks assume static, idealized settings. We present SVRPBench, the first open benchmark to capture high-fidelity stochastic dynamics in vehicle routing at urban scale. Spanning more than 500 instances with up to 1000 customers, it simulates realistic delivery conditions: time-dependent congestion, log-normal delays, probabilistic accidents, and empirically grounded time windows for residential and commercial clients. Our pipeline generates diverse, constraint-rich scenarios, including multi-depot and multi-vehicle setups. Benchmarking reveals that state-of-the-art RL solvers like POMO and AM degrade by over 20% under distributional shift, while classical and metaheuristic methods remain robust. To enable reproducible research, we release the dataset and evaluation suite. SVRPBench challenges the community to design solvers that generalize beyond synthetic assumptions and adapt to real-world uncertainty.

Paper Structure

This paper contains 40 sections, 16 equations, 14 figures, 11 tables.

Table of Contents

  1. Introduction
  2. The Case for a Realistic SVRP Benchmark.
  3. Our Contributions.
  4. Realistic Stochastic Modeling
  5. Time-Dependent Travel Time Modeling
  6. Customer Time Window Sampling
  7. Dataset Construction Pipeline
  8. Location Sampling.
  9. Demand Assignment.
  10. Time Window Assignment.
  11. Travel Time Matrix Construction.
  12. Constraint Integration.
  13. Validation.
  14. Various Scales.
  15. Evaluation Protocol
  16. ...and 25 more sections

Figures (14)

  • Figure 1: SVRPBench pipeline. The framework generates realistic SVRP instances through four stages: input generation, stochastic modeling, instance assembly, and evaluation with standardized metrics and solvers.
  • Figure 2: Comparison of real (top) and synthetic (bottom) routing instances across three cities.
  • Figure 3: Solver Comparison: Overall Performance Metrics.
  • Figure 4: Solver Performance by Problem Size.
  • Figure 5: CVRP 20 customers – Attention Model
  • ...and 9 more figures