WildCity Paper: May Mobility + NYU + Nvidia + Stanford Release First City-Scale Multimodal Dataset — 18 Autonomous-Fleet Trajectories Averaging 83.7km Each Enable Closed-Loop Urban Digital Twins

New arXiv paper introduces WildCity, a real-world multimodal city-scale dataset collected by autonomous fleets, with 18 trajectories averaging 83.7km each — preserving dynamic objects, lighting variations, and imperfect camera poses. Beyond the dataset, the team from May Mobility, NYU, Nvidia and Stanford establishes an urban-tailored reconstruction baseline, converts the reconstructions into a closed-loop simulator, and analyzes the three key challenges to simulation-ready urban digital twins: scalability, extrapolation, and uncertainty.