Manhattan Dial-a-Ride Benchmark Dataset with NYC TLC Taxi Trips, 2015–2016

This dataset provides the network and raw trip records for dynamic dial-a-ride (DARP) experiments on Manhattan, New York City. It includes (1) a custom virtual-stop network, (2) transformed yellow-taxi trip records for 24 days in 2015–2016, and (3) benchmark instance files for multiple service windows. Network. The study area is Manhattan, defined by the NYC Taxi and Limousine Commission (NYCTLC) taxi-zone boundaries. Virtual pickup and drop-off stop locations were derived from the OpenStreetMap (OSM) drivable road network using OSMnx: street intersections with degree ≥ 3 were extracted as candidate stops and thinned with a Poisson-disk filter at a minimum spacing of 150 m. The resulting stop set is provided in virtual_stops_latlon.geojson. The travel-time matrix between all stop pairs is provided in edge_time_matrix.txt (StartID, EndID, duration in seconds); this matrix was computed using OSRM based on OpenStreetMap data and post-processed to enforce the triangle inequality via Floyd–Warshall.Trip records. Raw taxi trip records were obtained from the NYCTLC open-data portal for 24 selected days across 2015 and 2016. Each day's records are stored as a JSON file ({date}_manhattan.json) containing fields: passenger count, pickup/drop-off coordinates, pickup/drop-off virtual stop IDs, and request timestamps.Benchmark instances. This folder contains solver-ready instances derived from the trip records above. Each instance corresponds to one day and one time window (e.g., a 2-hour or 4-hour slice starting at 7 AM or 11 AM) and is identified by date and window (e.g., 20150706_7-2h). Each instance comprises two files: TRIP_*.txt (request data with pickup/drop-off stop IDs and timestamps), and INSTANCE_*.txt (metadata: start time in seconds, number of requests, number of stops). Vehicle deployment files are provided separately for fleet sizes ranging from 500 to 3,000 vehicles.