New workflow generates HD maps using open geo-data, no sensors needed

High-definition maps are critical for self-driving cars, but traditional generation relies on costly sensor-equipped vehicles and high-precision reference data for quality checks. This dependency makes HD mapping expensive and inaccessible for many regions. A new paper from Ruidi He and colleagues proposes an engineering-oriented workflow that uses openly available geo-engineering datasets (like shapefiles) as the primary input. The system transforms this data through explicit intermediate representations into lane-level HD maps using the lanelet format. To verify map quality without external reference, the workflow integrates executable constraint-based checks derived from automated driving specs and road-design guidelines, detecting geometric, topological, and elevation inconsistencies.

Evaluated on real-world road-network data from four cities in Lower Saxony, Germany, the generated maps satisfied all selected constraints. In a controlled defect-injection study, the verification system detected every injected defect type with no false positives. The results demonstrate that geo-data-driven HD map generation with integrated verification can serve as a modular, inspectable complement to sensor-intensive approaches, especially when specialized measurement data or reference maps are unavailable. This could dramatically lower the barrier for deploying HD maps in new regions and enable faster scaling of automated driving systems.