The hard part of million-stop routing was not the route optimizer
A new technical report presents an architecture for optimizing last-mile routing at a large scale. The study demonstrates significant improvements in routing efficiency, processing one million stops in approximately 20 minutes on standard hardware. This approach aims to make large routing workloads manageable without the need for extensive infrastructure or manual pre-partitioning.
- ▪The report achieved 23.3% less measured distance compared to Amazon's baseline routes.
- ▪It processed one million stops in about 20 minutes using commodity hardware.
- ▪The architecture focuses on organizing routing into bounded, composable stages.
Opening excerpt (first ~120 words) tap to expand
Rethinking Last-Mile Routing at Scale Near-linear planning on commodity hardware This repository contains a technical report describing a practical architecture for large-scale last-mile route optimization. The core idea: at million-stop scale, vehicle routing stops being only an optimization problem and becomes a systems problem involving partitioning, boundary repair, graph reuse, bounded route-level optimization, and orchestration. Paper 📄 Download the PDF Main results Under a shared external measurement protocol based on OSRM and Google Maps, using the public Amazon Last Mile Routing Research Challenge dataset: 23.3% less measured distance relative to the released Amazon baseline routes 11.1% fewer routes 17.59% mean depot-level distance reduction 1,000,000 stops processed in ~20…
Excerpt limited to ~120 words for fair-use compliance. The full article is at GitHub.