Poryos2026 (VRPTW)

Poryos2026 is the generated benchmark family introduced by MAMUT-routing. It's VRPTW family is derived from the generated CVRP layer. This split is deliberate: first generate a real-world CVRP customer set and travel model from OSM data, then add service times and time windows under an explicit VRPTW generation contract. This avoids mixing geography, travel-time semantics, and temporal constraints into an opaque one-step generator.

The current VRPTW layer uses integer arc costs for reproducibility. The main metric variant is fastest, where arc costs are directly interpreted as travel times from the OSM road-network model. Earlier design notes also considered a euclidean VRPTW variant obtained by converting Euclidean meters to travel time using a fixed average speed of 14 m/s (around 50 km/h). We also tested another conversion consisting in calculating the ratio between fastest and another non-time-based metric such as shortest or euclidean on the CVRP layer, then applying that ratio to the euclidean distances to get a time-based arc-cost matrix. Both those variants are highly synthetic and less directly connected to real-world travel times than the fastest variant. They also add another layer of complexity to the generation process, while being less useful for studying real-world VRPTW behavior. For these reasons, we decided to simplify the current VRPTW release by only including the meaningful fastest instances.

Time windows are generated synthetically but with inspiration from Solomon's original benchmark logic. The workbench now supports two methods:

• route_centered: a Solomon C-class inspired method. A nearest-neighbour reference route is constructed, customer arrival times are simulated, and each time window is centered around the corresponding arrival time with randomized width. This is suited to clustered or route-structured temporal patterns.
• reachable_interval: a Solomon R/RC-class inspired method. Each customer receives a randomized window whose center is chosen inside an interval compatible with reaching that customer from the depot and returning before the depot horizon closes.

In both methods, service times are sampled deterministically from seeded random parameters, the depot horizon is explicit, and each customer time window is repaired if necessary so that the elementary route depot -> customer -> depot is always feasible. This elementary-route guarantee is weaker than providing a known full feasible solution, but it prevents trivially invalid customers and gives solvers a feasible upper-bound fallback when the fleet is unlimited or sufficiently loose.

The current generated time-window metadata records the method, horizon, service-time ratio, time-window ratio, and repair count. This is important for external users: a VRPTW instance is not fully described by coordinates, demands, service times, and time windows alone. Its benchmark meaning also depends on how those time windows were generated, whether they are route-centered or independently reachable, and how feasibility repairs were applied.

The VRPTW Poryos2026 family is meant to complement, not replace, the historical Sintef2008, Dimacs2021, and Ortec2022 families. Sintef2008 and Dimacs2021 provide continuity with decades of classical Solomon-style research. Ortec2022 provides a realistic industry-derived benchmark. Poryos2026 adds an open generation toolchain where geography, travel model, temporal policy, objective convention, and visualization are all explicit and reproducible.

Licensing note: because this family is derived from the OSM-based Poryos2026 CVRP layer, the OSM-derived VRPTW instances, sidecars, route-rendering artifacts, and related benchmark data are distributed under the Open Data Commons Open Database License (ODbL) v1.0 where applicable, with attribution to OpenStreetMap and its contributors.

The current family contains instances alongside heuristic BKS for both classical objectives: a MonoCost BKS and a HierarchicalVehicleCost BKS.

Data

Instances and best-known solutions: /benchmarks/vrptw/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.