Study audits CityGML-to-USD conversion in urban digital twins, suggesting improvements for semantic mapping.
Urban digital twin platforms are converging on OpenUSD as their scene backbone, and Japan’s PLATEAU programme publishes CityGML 2.0 city models as open data nationwide. Yet OpenUSD has no ratified schema for coordinate reference systems, no standard mapping for CityGML semantics, and stores mesh vertices as 32-bit floats. What survives conversion into this format has not been measured. We audit five PLATEAU LOD2 building tiles (6,888 buildings, central Tokyo and regional Kaga) along two routes: the reference open-source toolchain (citygml-tools, cjio, Blender) and a semantics-aware direct writer built as an existence proof. The toolchain route preserved zero of 19,857 semantic attribute instances and no surface-type structure. It stored latitude and longitude degrees as Cartesian coordinates under a declared unit of metres, and its axis handling left longitude standing as the vertical axis in world space. USD can neither represent nor detect either error. The direct writer preserved 100% of the same attributes using only standard USD mechanisms. A controlled experiment isolates the float32 floor across coordinate-handling choices, with a vertex-level check confirming the toolchain’s stored coordinates are exactly the float32 image of the source degrees. Worst-case vertex error spans four orders of magnitude: 715-721 mm with degrees stored as-is, 2.0-4.4 mm projected (growing with distance from the projection origin), 0.06-0.09 mm under a local anchor. Semantic queries fail entirely on the toolchain route, yet geometric error becomes negligible once coordinates are localized. The ceiling is semantic and declarative rather than numeric, and evidence gathered now can still shape the georeferencing schema the Alliance for OpenUSD has yet to deliver.
No takes yet. Share an insight, caveat, or question.
Jiawei Jing (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: