Poryos2026 (TDVRPTW)

The time-dependent layer of the workbench-generated Poryos2026 family consists of TDVRPTW instances whose congestion is derived from real city road networks rather than from the classic category-times-profile constructions of the literature (for that lineage see Dabia2013 and Lera2026). Depot and customers are sampled by point-of-interest or hybrid sampling on the OpenStreetMap road graph of a real city: Lyon, Paris, San-Francisco, Hong-Kong, or Tokyo (central extracts). Every edge of the network carries an hourly speed profile over a 24-hour day. Two synthetic traffic models produce those profiles, and both ship as instances (the model tag is part of every instance name): bpr simulates a commuter population whose morning, lunch and evening trips load the network, converting hourly flows to speeds through the Bureau of Public Roads volume-delay function with road-class capacities; wave imposes a bimodal rush-hour speed dip scaled by road class, distance to the city center and a seeded per-edge jitter. Each model comes in three intensities (light, moderate, heavy), so rush-hour congestion emerges from the network structure. Under the heavy commuter scenario, the median arc takes about three times longer at 08:00 than at 03:00, while side streets mostly keep free flow and the best path between two customers can change with the departure time.

Per-arc travel times are the exact result of driving the fastest free-flow path through the time-varying edge speeds. They are stored as a compact road-graph specification rather than as explicit arrival-time-function files. The specification contains the trimmed subgraph of the city network that the instance actually uses, together with its speed profiles. The arrival-time functions are rebuilt deterministically whenever an instance is loaded (a pinned shortest-path rule fixes one canonical path per arc, exact per-edge arrival functions are sampled exactly along it, and the samples are decimated with a recorded tolerance), and a recorded SHA-256 fingerprint guarantees that every rebuild reproduces the canonical functions bit for bit. Time windows are synthesized by a time-dependent variant of the workbench's route-centered method and repaired to time-dependent feasibility, so every customer is individually serveable within the day. Sizes range from 10 to 1000 customers; the small sizes are exact-solver-friendly, the large ones stress heuristics on genuinely road-network-shaped time dependence.

Licensing note: the road networks derive from OpenStreetMap, and the benchmark data of this family is distributed under ODbL 1.0 with attribution to OpenStreetMap and its contributors.

The family contains 360 instances over 6 sizes (10 to 1000 customers), 5 cities, 2 traffic models, 3 intensities and 2 sampling methods. Every instance carries a Duration BKS, seeded and re-seeded by KAYROS anytime campaigns on Grid5000 through the standard improve-only, checker-authoritative pipeline.

Data

Instances and best-known solutions: /benchmarks/tdvrptw/poryos2026/ on the website. Repository: https://github.com/ANR-MAMUT/MAMUT-routing-Poryos2026.

This family is one problem-type layer of the Poryos2026 collection; the collection README (layout, conventions, sidecars) is on the collection page.