Valdrada

Valdrada is a digital twin of New York City developed and maintained by Mehmet Kerem Turkcan. The city is compiled from public records: NYC Building Footprints joined to PLUTO and facade-inspection filings, the Street Centerline file for streets, lanes and speed limits, OpenStreetMap to fill gaps in footprints, and street-furniture records for trees, hydrants, bus shelters, kiosks and subway entrances. The compiled city covers all five boroughs with 1,085,937 buildings in 3,532 near tiles and 261 far-field tiles; Central Park (with 16,951 trees in 33 species groups), 125th Street, Times Square and Midtown are modelled in extra detail.

On top of the model sit a WebGL2 renderer, lane-level traffic (IDM car following with signal phases and turn planning), signal-aware pedestrians on a sidewalk graph, a New York vehicle fleet and two rail lines. Sensors return RGB, Cityscapes-compatible semantic and instance segmentation, occlusion and 3D pose, and metric depth. The stated uses are synthetic data generation, closed-loop scenario simulation and streetscape studies; recorded takes export to OpenUSD for Blender and Unreal Engine 5.

It runs as a simulation server with a TCP API and a Python client modelled on CARLA, either from a Windows x64 binary or built from source with npm, with a browser demo on Hugging Face. Code is MIT, documentation CC BY 4.0, and the compiled city ODbL 1.0 because it contains OpenStreetMap data; the full dataset is about 7.6 GB. Some pedestrian assets carry restrictions on dataset and AI-training use, and the README points to a procedural-pedestrian mode for those cases.

Status: released, version 0.4.0, 2026. The earlier repository and Hugging Face names "boundless-nyc" now redirect to Valdrada; the pages read do not mention the old name.