Rifki2020 (TDVRP)¶
The TDVRP layer of Rifki2020 is the same 180-instance family with the time windows removed: same real Lyon envelope-FIFO arrival-time functions (shared bit-for-bit with the TDVRPTW layer), same curated demands, capacities, fleet sizes and service times, same Duration checker contract. It combines real-world time dependency, including genuine vertical steps in the arrival-time functions, with the unpruned search spaces of the no-time-window setting. No previously published solutions exist for this variant; all BKS come from KAYROS heuristic runs on Grid5000.
Licensing note: as for the TDVRPTW layer, MAMUT-authored curation artifacts are under the MIT License, while underlying third-party data is not relicensed.
The family currently contains 180 instances over 6 sizes, each with a Duration BKS.
Data
Instances and best-known solutions: /benchmarks/tdvrp/rifki2020/ on the website. Repository: https://github.com/ANR-MAMUT/MAMUT-routing-TDVRP-Rifki2020.
Repository README¶
Rifki2020 TDVRP benchmark family (TW-free twin, real Lyon road-network travel times)¶
Satellite benchmark repository of MAMUT-routing (ANR MAMUT, ANR-22-CE22-0016), mounted there as benchmarks/TDVRP/Rifki2020. Self-contained: instances, canonical arrival-time-function (ATF) sidecars and best-known solutions ship together.
This is the TDVRP twin of MAMUT-routing-TDVRPTW-Rifki2020: the same 180 Duration-Minimization instances (6 sizes, 10–60 customers, × 30 ids, named Rifki-<id>) with the time windows removed — capacity, fleet size, service times, demands, coordinates and the time-dependent travel times are identical, and the ATF sidecars are byte-identical to the TDVRPTW twin's (satellites are self-contained, so they are duplicated here rather than referenced).
Everything else — raw source (Rifki, Chiabaut & Solnon 2020 Lyon data, σ = 100, K = 60 steps of 720 s), the choice of the K = 60 granularity over the finest K = 120 (halved breakpoint counts in the same weight class as the other TD families; the granularities are independent published aggregations, so this is a choice of discretization, not a compression), the arrival-time lower-envelope FIFO restoration (exact MD92, NDCPWLF vertical steps at surviving upward jumps), the deliberate replacement of the original [RCS20] time-window files by Onyr's legacy generated attributes, the deterministic stress-layout coordinates, the all-gzipped sidecar storage, and the "checker defines the objective" rule — is documented in the TDVRPTW twin's README and applies verbatim; per-instance provenance lives in each instance's metadata.
Population pipeline: populate_td_rifki2020 v2. Underlying travel-time data: Rifki, Chiabaut & Solnon (2020); MAMUT-authored curation artifacts under the MIT license; this notice does not relicense the underlying third-party data.