RoomPlan and ARKit already provide most of the components required for professional as-built verification: semantic room reconstruction, LiDAR scene meshes, world tracking, relocalization, reference-object tracking, and on-device ML.
What is still missing is a system-level, hardware-accelerated API that compares a reference design with the live reconstructed physical environment.
The proposed API should accept reference geometry or a semantic spatial model, persistent object identifiers, a reference/world transform, and configurable dimensional tolerances, and continuously return measured deviations between the design model and the observed scene.
Example outputs:
Wall W103
Endpoint deviation: -31 mm
Status: Out of tolerance
Pipe P204
Centerline deviation: +12 mm
Status: Within tolerance
Valve V204
Status: Expected but not observed
Confidence: 0.93
Door D14
Status: Matched
Deviation: 6 mm
The API should expose measured deviation, confidence, tracking quality, and estimated measurement accuracy.
The important capability is not support for another CAD or BIM file format. Third-party applications can convert IFC, STEP, USD, USDZ, or other engineering formats into the reference representation required by the framework.
The important system-level workflow is:
Design intent → Spatial registration → Live LiDAR / camera reconstruction → Reference-to-reality comparison → Quantified deviations
This would allow an engineer to walk through a construction site or industrial plant with an iPhone, iPad Pro, or Vision Pro and receive real-time verification such as:
this wall is positioned correctly
this wall is 31 mm too short
this pipe is within installation tolerance
this valve is missing
this opening is too small
this equipment is installed at the wrong elevation
RoomPlan currently answers:
“What is here?”
A comparison API would allow Apple spatial frameworks to also answer:
“How does what is here differ from what should be here?”
This could support construction QA/QC, BIM as-built verification, renovation, industrial installation, commissioning, digital twins, facility management, and automated progress inspection.
Developers can already implement custom mesh comparison algorithms themselves, but this requires recreating functionality that could be much more accurate and efficient when implemented close to Apple’s calibrated camera, LiDAR, IMU, Neural Engine, and ARKit tracking stack.
A native reference-to-reality comparator would therefore be much more valuable than simply adding additional export formats.
Topic:
ARKit
SubTopic:
ARKit Q&A