Real-World Vehicle Routing Using Adaptive Large Neighborhood Search

Abstract

Our work addresses a real-world freight transportation problem with a broad set of characteristics. We build upon the classical work of Ropke and Pisinger [10] and propose an effective realization of the adaptive large neighborhood search (ALNS) with constant time complexity for a large portion of frequent steps in insertion and removal heuristics at the cost of additional pre-calculations. Our minimization process handles different objectives with cost models of heterogeneous vehicles. We demonstrate the generic applicability of the proposed solver on various vehicle routing problems. With the help of the standard Li & Lim benchmarks [6] for pickup and delivery with time windows, we show its capabilities compared to the best-found solutions and the original ALNS. Experiments on real-world delivery routing problems provide a comparison with the original implementation by the company Wereldo in OR-Tools [8], where we achieve significant cost savings, faster runtime, and memory savings by order of magnitude. Performance on large-scale real-world instances with more than 300 vehicles and 1,200 pickup and delivery requests is also presented, achieving less than an hour runtimes.