Developing an Urban Road Dataset: A Multi-Sensor Framework for DT and AI-Based Road Infrastructure Management

High-quality multi-sensor data are essential for advancing Digital Twins (DTs), AI-based road infrastructure monitoring, and automated urban modelling. Despite growing interest, publicly available datasets that simultaneously integrate mobile mapping systems (MMS), aerial LiDAR (ALS), imagery, BIM-derived assets, and detailed pavement defect annotations within a unified, DT-ready framework remain scarce. This paper presents the Turin Urban Road Dataset, a multi-sensor, multi-layer data framework consolidating research outputs developed within the Turin Digital Twin initiative. The dataset emerges from the integration of previously validated methodologies for pavement condition assessment, BIM-based asset modelling, and point cloud semantic classification, formalised here into a coherent and reusable resource for urban road infrastructure research. It combines high-density MMS LiDAR (1,100 pts/m² on pavement), ALS LiDAR (30–40 pts/m²), RGB/NIR and 360° panoramic imagery, georeferenced BIM entities, and over 7,000 manually classified pavement defects across three representative urban scenes totalling approximately 200 million annotated points. All data sources are harmonised in EPSG:25832 through a quality-controlled pipeline ensuring geometric alignment, semantic coherence, and metadata completeness, consistent with DT data-quality principles. Rather than a fully validated benchmark, the dataset represents a structured and reproducible foundation for future experimentation in semantic segmentation, AI-based defect detection, and DT development, bridging the gap between geospatial acquisition, semantic enrichment, and infrastructure-level decision support.