INFORMS RAS 2026 Problem Solving Competition

Railroad Blocking Problem for North American Class I Railroads — Data Mirror Site

Map of major U.S. Class I railroads, intermodal terminals, and RoRo ports (illustrative)

Header image: “US railway map” by Wikideas1, Wikimedia Commons, CC0 1.0 (public domain); own work created with QGIS using U.S. Bureau of Transportation Statistics data. Shown here as an illustrative overview of the Class I railroad network, intermodal terminals, and RoRo ports; it is not the benchmark network.

This is the data mirror site maintained by the ASU Trans+AI Lab for the INFORMS Railway Applications Section (RAS) 2026 Problem Solving Competition. It mirrors the benchmark instances, schemas, validators, scoring tools, and documentation so the dataset stays citable and reproducible. It is not the official competition site: the Kaggle and INFORMS RAS pages linked below are the authoritative sources for the competition itself.

Kaggle competition page INFORMS RAS competition page GitHub repository

Official announcements

Please cite

Peiheng Li and Xuesong (Simon) Zhou. INFORMS RAS 2026 Problem Solving Competition. Kaggle, 2026. https://www.kaggle.com/competitions/informs-ras-2026-problem-solving-competition

@misc{li_zhou_ras2026,
  author       = {Li, Peiheng and Zhou, Xuesong (Simon)},
  title        = {INFORMS RAS 2026 Problem Solving Competition},
  year         = {2026},
  howpublished = {Kaggle},
  url          = {https://www.kaggle.com/competitions/informs-ras-2026-problem-solving-competition}
}

Users relying on upstream public source datasets should additionally cite the applicable original sources, including FAF and OpenStreetMap, as appropriate.

Research benchmark and data provenance notice

RAS2026-PSC is a research-derived and benchmark-calibrated competition dataset created for the INFORMS RAS 2026 Problem Solving Competition. It is not an operational or proprietary railroad dataset. Demand records were developed using transformed Freight Analysis Framework (FAF)-based freight-flow references; users seeking original FAF data should obtain it directly from the official BTS/FAF sources. Physical network connectivity is research-derived in part from OpenStreetMap and additional competition-specific research curation. Applicable OpenStreetMap attribution and ODbL terms remain with OSM-derived components. Benchmark yard definitions, capacities, costs, and other parameters are research abstractions for algorithmic evaluation and should not be interpreted as current railroad operating data. Participants and users take full responsibility and liability for their use of the materials.

Full notice: DATA_PROVENANCE_AND_USE.md · Notices: NOTICE.md

The problem in brief

Railcars are consolidated into blocks at classification yards and travel intact between yards. Given a physical rail network and yard-to-yard commodity demand, participants build a zero-based blocking plan: which blocks to open, the single sequence of blocks each commodity follows, and the physical route of each block. Plans are checked for feasibility (constraints C1–C8) and ranked by total cost (block fixed, transportation, classification handling, interchange), with a volume-weighted Stress Score when full feasibility is unattainable. Three network layers (L1, L2, L3) × three demand multipliers (0.5×, 1.0×, 2.0×) form the nine official cases.

Problem statement (PDF)

Downloads

Full package — GitHub Release v2.1 Repository ZIP (compressed inputs) SHA-256 checksums

Item Location
Per-layer inputs (node.csv, demand.csv, setting.csv, yard_car_demand_summary.csv, link.csv.zip) datasets/l1/, datasets/l2/, datasets/l3/
Yard-to-yard distance matrix datasets/yard_to_yard_min_distance.csv.zip
JSON schemas and CSV validator datasets/schemas/
Validator, metric notebook, packing scripts, sample solutions scoring/
Field reference datasets/DATASET_README.md
Scoring reference scoring/SCORE_README.md

Quick start:

git clone https://github.com/asu-trans-ai-lab/RAS2026-PSC.git
cd RAS2026-PSC
pip install -r requirements.txt
python unpack.py
python datasets/schemas/validate_csvs.py
cd scoring
python fast_validator_v2_0.py solution_result_l1_10.json --od-matrix od_distance_matrix.csv

Organization team

Institutional affiliations are provided to identify the competition leadership and problem-development roles. Their listing should not be interpreted as a separate institutional license, sponsorship statement, operational endorsement, or transfer of ownership of the released benchmark.

Judging committee

Name Organization
Cynthia Barnhart MIT
Ravi Ahuja Optym
David Hunt Oliver Wyman
Michael Hewitt Loyola University Chicago
Gunnar Feldmann Norfolk Southern
Edward Lin Independent / retired Norfolk Southern
Clark Cheng RailTek
Marc Meketon Independent
John Fuller Union Pacific
Carl Van Dyke CVD Zone
Baoyu Zhou Arizona State University

Affiliations are listed for identification only.

Upstream sources

Freight Analysis Framework (FAF) — U.S. Bureau of Transportation Statistics (BTS); FAF5 developed with support from the Federal Highway Administration (FHWA).

FAF5 is listed because it belongs to the provenance family used in developing the RAS 2026 benchmark. The current BTS FAF portal may contain newer FAF generations; do not assume a later FAF release was used unless explicitly documented.

OpenStreetMap (OSM) — © OpenStreetMap contributors, Open Database License (ODbL) 1.0.

Licenses and notices

Project-authored code is MIT-licensed (LICENSE-CODE); the code license does not extend to the datasets or third-party-derived data, and no blanket data license is asserted. See DATA_PROVENANCE_AND_USE.md and NOTICE.md.

RAS2026-PSC is an openly accessible research benchmark archive for reproducible railroad-blocking algorithm development. It integrates transformed public freight-flow references, open spatial/network references, and competition-specific research curation, while preserving the attribution and use conditions of the underlying sources.