ULTRA: UnLimited TRAnsfers for Efficient Multimodal Journey Planning

ULTRA is a C++ framework for efficient journey planning in multimodal networks that combine public transit and non-schedule-based transfer modes (e.g., walking, cycling, e-scooter). It was developed at KIT in the group of Prof. Dorothea Wagner. This repository contains code for the following publications:

• UnLimited TRAnsfers for Multi-Modal Route Planning: An Efficient Solution. Moritz Baum, Valentin Buchhold, Jonas Sauer, Dorothea Wagner, Tobias Zündorf. In: Proceedings of the 27th Annual European Symposium on Algorithms (ESA'19), Leibniz International Proceedings in Informatics, pages 14:1–14:16, 2019. pdf arXiv

• Integrating ULTRA and Trip-Based Routing. Jonas Sauer, Dorothea Wagner, Tobias Zündorf. In: Proceedings of the 20th Symposium on Algorithmic Approaches for Transportation Modelling, Optimization, and Systems (ATMOS'20), OpenAccess Series in Informatics, pages 4:1–4:15, 2020. pdf

• An Efficient Solution for One-to-Many Multi-Modal Journey Planning. Jonas Sauer, Dorothea Wagner, Tobias Zündorf. In: Proceedings of the 20th Symposium on Algorithmic Approaches for Transportation Modelling, Optimization, and Systems (ATMOS'20), OpenAccess Series in Informatics, pages 1:1–1:15, 2020. pdf

• Fast Multimodal Journey Planning for Three Criteria. Moritz Potthoff, Jonas Sauer. In: Proceedings of the 24th Workshop on Algorithm Engineering and Experiments (ALENEX'22), SIAM, pages 145–157, 2022. pdf arXiv

• Efficient Algorithms for Fully Multimodal Journey Planning. Moritz Potthoff, Jonas Sauer. In: Proceedings of the 22nd Symposium on Algorithmic Approaches for Transportation Modelling, Optimization, and Systems (ATMOS'22), OpenAccess Series in Informatics, pages 14:1–14:15, 2022. pdf

• ULTRA: Unlimited Transfers for Efficient Multimodal Journey Planning. Moritz Baum, Valentin Buchhold, Jonas Sauer, Dorothea Wagner, Tobias Zündorf. In: Transportation Science, volume 57(6), pages 1536-1559, 2023. pdf

• Fast and Delay-Robust Multimodal Journey Planning. Dominik Bez, Jonas Sauer. In: Proceedings of the 26th Workshop on Algorithm Engineering and Experiments (ALENEX'24), SIAM, pages 105–117, 2024. pdf arXiv

• Closing the Performance Gap Between Public Transit and Multimodal Journey Planning. Jonas Sauer. PhD thesis, 2024. pdf

Compiling

To compile all executables in release mode, run

mkdir -p cmake-build-release
cd cmake-build-release
cmake .. -DCMAKE_BUILD_TYPE=Release && cmake --build . --target All --config Release

Make sure you have OpenMP installed.

Usage

Most preprocessing steps and query algorithms are provided in the console application ULTRA, which offers the following commands:

• Contraction Hierarchies (CH) computation:
  • buildCH performs a regular CH precomputation. The output is used by the (Mc)ULTRA query algorithms for the Bucket-CH searches.
  • buildCoreCH performs a Core-CH precomputation. The output is used by the (Mc)ULTRA shortcut computation and by the MCSA and M(C)R query algorithms.
• (Mc)ULTRA shortcut computation:
  • computeStopToStopShortcuts computes stop-to-stop ULTRA shortcuts for use with ULTRA-CSA and ULTRA-RAPTOR.
  • computeEventToEventShortcuts computes event-to-event ULTRA shortcuts for use with ULTRA-TB.
  • computeDelayEventToEventShortcuts computes delay-tolerate event-to-event ULTRA shortcuts.
  • computeMcStopToStopShortcuts computes stop-to-stop McULTRA shortcuts for use with ULTRA-McRAPTOR and UBM-RAPTOR.
  • computeMcEventToEventShortcuts computes event-to-event McULTRA shortcuts for use with ULTRA-McTB and UBM-TB.
  • augmentTripBasedShortcuts performs the shortcut augmentation step that is required for UBM-TB.
  • validateStopToStopShortcuts and validateEventToEventShortcuts test the validity of the computed shortcuts by comparing them to paths in the original transfer graph.
• Original TB transfer generation:
  • raptorToTripBased takes a network in RAPTOR format as input and runs the TB transfer generation.
  • With a transitively closed transfer graph as input, this performs the original TB preprocessing.
  • With stop-to-stop ULTRA shortcuts as input, this performs the sequential ULTRA-TB preprocessing.
  • The parameter "Route-based pruning?" enables the optimized preprocessing proposed by Lehoux and Loiodice.
• Query algorithms:

Networks

We use custom data formats for loading the public transit network and the transfer graph: The Intermediate format allows for easy network manipulation, while the RAPTOR format is required by the preprocessing and all query algorithms except for CSA, which uses its own format. The Switzerland and London networks used in our experiments are available at https://i11www.iti.kit.edu/PublicTransitData/ULTRA/ in the required formats. Unfortunately, we cannot provide the Germany and Stuttgart networks because they are proprietary.

The Network application provides commands for manipulating the network data and for converting public transit data to our custom format. It includes the following commands:

• parseGTFS converts GFTS data in CSV format to a binary format.
• gtfsToIntermediate converts GFTS binary data to the Intermediate network format.
• intermediateToCSA converts a network in Intermediate format to CSA format.
• intermediateToRAPTOR converts a network in Intermediate format to RAPTOR format.
• loadDimacsGraph converts a graph in the format used by the 9th DIMACS Implementation Challenge to our custom binary graph format.
• duplicateTrips duplicates all trips in the network and shifts them by a specified time offset. This is used to extend networks that only comprise a single day to two days, in order to allow for overnight journeys.
• addGraph adds a transfer graph to a network in Intermediate format. Existing transfer edges in the network are preserved.
• replaceGraph replaces the transfer graph of a network with a specified transfer graph.
• reduceGraph contracts all vertices with degree less than 3 in the transfer graph.
• reduceToMaximumConnectedComponent reduces a network to its largest connected component.
• applyBoundingBox removes all parts of a network that lie outside a predefined bounding box.
• applyCustomBoundingBox removes all parts of a network that lie outside a specified bounding box.
• makeOneHopTransfers computes one-hop transfers for all stops whose distance is below a specified threshold. This is used to create a transitively closed network for comparison with non-multi-modal algorithms.
• applyMaxTransferSpeed applies a maximum transfer speed to all edges in the transfer graph.
• applyConstantTransferSpeed applies a constant transfer speed to all edges in the transfer graph and computes the travel times accordingly.

An example script that combines all steps necessary to load a public transit network is provided at Runnables/BuildNetworkExample.script. It can be run from the Network application using runScript BuildNetworkExample.script. It takes as input GFTS data in CSV format located at Networks/Switzerland/GTFS/ and a road graph in DIMACS format located at Networks/Switzerland/OSM/dimacs.

Multiple Transfer Modes

The algorithms listed above support bimodal networks with public transit and a single transfer mode. Additionally, this framework provides algorithms for multimodal networks with multiple transfer modes. The required multimodal data structures can be built with the following commands in Network:

• buildMultimodalRAPTORData converts unimodal RAPTOR data into multimodal RAPTOR data. The transfer graph contained in the RAPTOR data is used for the "free" transfers whose transfer time is not penalized. The transfer graphs for the non-"free" modes must be added separately with the addModeToMultimodalRAPTORData.
• addModeToMultimodalRAPTORData adds a transfer graph for a specified transfer mode to the given multimodal RAPTOR data.
• buildMultimodalTripBasedData converts unimodal TB data into multimodal TB data. The transfer graph contained in the TB data is used for the "free" transfers whose transfer time is not penalized. The transfer graphs for the non-"free" modes must be added separately with the addModeToMultimodalTripBasedData.
• addModeToMultimodalTripBasedData adds a shortcut graph for a specified transfer mode to the given multimodal TB data. ` Additionally, the command buildFreeTransferGraph in ULTRA builds a "free" transfer graph by connecting all pairs of stops within a specified geographical distance and then computing the transitive closure.

ULTRA shortcuts for networks with multiple transfer modes can be computed with the following commands in ULTRA:

• computeMultimodalMcStopToStopShortcuts computes multimodal stop-to-stop McULTRA shortcuts for use with ULTRA-McRAPTOR and UBM-RAPTOR.
• computeMultimodalMcEventToEventShortcuts computes multimodal event-to-event McULTRA shortcuts for use with UBM-HydRA.

The ULTRA application offers the following query algorithms. All algorithms optimize arrival time, number of trips and one transfer time criterion per transfer mode.

• runMultimodalMCRQueries: MCR for full Pareto sets
• runMultimodalULTRAMcRAPTORQueries: ULTRA-McRAPTOR with stop-to-stop shortcuts for full Pareto sets
• runMultimodalUBMRAPTORQueries: UBM-RAPTOR with stop-to-stop shortcuts for restricted Pareto sets
• runMultimodalUBMHydRAQueries: UBM-HydRA with event-to-event shortcuts for restricted Pareto sets

One-to-Many Journey Planning

The query algorithms in the ULTRA application only support one-to-one queries. The ULTRAPHAST application provides algorithms for one-to-all and one-to-many queries:

Random ball target sets can be generated with the command createBallTargetSets. CH and Core-CH precomputations for these target sets can be run with buildUPCHForTargetSets and buildCoreCHForTargetSets, respectively.

Delay-Robustness

The application DelayExperiments provides commands for evaluating Delay-ULTRA, the variant of ULTRA that anticipates possible vehicle delays. The delay-robust shortcut computation itself is run with the command computeDelayEventToEventShortcuts in ULTRA. All delays up to the specified limit (measured in seconds) are accounted for. DelayExperiments provides the following commands:

• GenerateDelayScenario generates a delay scenario for the given network, using a synthetic delay model.
• GenerateDelayQueries generates queries for the specified delay scenario that are answered incorrectly by an algorithm without delay information.
• BuildFakeDelayData takes as input a network with regular ULTRA shortcuts and converts it to the format used by Delay-ULTRA. This is useful for comparing Delay-ULTRA to regular ULTRA.
• RunDelayUpdatesWithoutReplacement simulates basic delay updates for the given delay scenario.
• RunDelayUpdatesWithReplacement simulates advanced delay updates for the given delay scenario. A heuristic replacement search is performed to find missing shortcuts.
• MeasureDelayULTRAQueryCoverage measures the result quality of TB using Delay-ULTRA shortcuts.
• MeasureHypotheticalDelayULTRAQueryCoverage measures the result quality of TB using Delay-ULTRA shortcuts, assuming that updates can be performed instantly.
• MeasureDelayULTRAQueryPerformance measures the query performance of TB using Delay-ULTRA shortcuts.