Recent breakthroughs in Scan to BIM

The short version

The bottleneck in Scan to BIM has always been the modeling step, not the scanning. The developments below all attack that bottleneck from different directions: better machine learning on 3D data, new intermediate representations between point cloud and BIM, faster capture, and open standards that let models move between design and simulation tools.

1. Machine learning on point clouds got much better

Research on 3D deep learning has advanced quickly. Point Transformer V3 (CVPR 2024) replaced expensive neighbour searches with a serialized point ordering, which its authors report made it about three times faster with ten times lower memory than its predecessor while enlarging the area each point can "see". In 2025, Meta's Sonata showed self-supervised pre-training for these models, selected as a highlight at CVPR 2025. Pre-training matters for Scan to BIM because labelled building scans are scarce: models can learn general 3D structure from unlabelled data, then be fine-tuned on walls, pipes and equipment.

The practical effect is that segmenting a building scan into elements (what is wall, what is duct, what is pump) is now reliable enough to automate large parts of modeling, with people reviewing rather than tracing. How AI is automating Scan to BIM goes deeper.

2. Meshes became a real bridge into Revit

Revit has always linked point clouds as reference only. Autodesk's ReCap Pro 2026 added locally generated segmented meshes (.rcmr) that a Revit plug-in can place into a project as families, which Autodesk says significantly reduces manual modeling. A mesh is not a parametric wall, but it is a big step from "dots to trace" towards "objects to refine". See how to convert a point cloud to Revit and IFC or Revit?.

3. Gaussian splats arrived in AEC

Gaussian splatting, a technique that represents a scene as millions of small coloured blobs, renders photorealistic views of a space quickly. AEC Magazine describes splats as a new tier in the reality-capture stack rather than a replacement for laser scanning. Scanner makers have started pairing LiDAR with splats: XGRIDS' LCC for Revit brings SLAM-based splats into Revit, with LiDAR providing the accuracy and splats the visual context. For BIM teams the value is context: a photoreal view next to the model makes it easier to check what an object really is.

4. Mobile scanning closed much of the accuracy gap

Mobile and backpack SLAM scanners capture a building many times faster than a tripod scanner. A 2024 peer-reviewed study in Remote Sensing tested NavVis VLX and Leica BLK2GO against a total station and a static scanner, and reported indoor errors around 5 mm, with larger errors outdoors. That is good enough for many LOD 300 existing-conditions models, which is why mobile capture is now common on large interiors. Choosing a scanner for Scan to BIM compares the options.

5. Phones became credible for small jobs

Phone and tablet LiDAR is not survey grade, but studies show it is usable for some BIM work. A 2024 study comparing iPhone LiDAR with a FARO scanner found most measured elements within centimetres. Our public samples include two sites rebuilt from Niantic Spatial phone captures. See can you do Scan to BIM with an iPhone? for where it works and where it does not.

6. The open standards matured

• IFC 4.3 became an ISO standard. buildingSMART announced in 2024 that IFC 4.3 was published as ISO 16739-1:2024, extending IFC to infrastructure. That strengthens IFC as the long-term record format for Scan to BIM deliverables.
• OpenUSD reached Core Specification 1.0. The Alliance for OpenUSD released Core Specification 1.0 in December 2025, the first formal specification of the format used for digital twins and simulation.
• The LOD specification kept improving. BIMForum published its 2023 LOD Specification (final files February 2024) focused on consistency and readability, and a 2025 edition followed.

7. Scan to BIM now feeds simulation

NVIDIA's work on SimReady assets and OpenUSD has pushed the idea that a building model should also be a physics-ready simulation asset. The same separated objects that make a good BIM model, once given collision shapes, materials and moving parts, let robots be trained and tested in a twin of the real site. From Scan to BIM to SimReady explains the extra steps.

What it means for your next project

• Expect days, not weeks. Automated modeling makes 48-hour delivery realistic for many sites.
• Still specify LOD and LOA. Automation does not remove the need for a clear brief. See how to write a Scan to BIM specification.
• Ask for every object separated. Equipment and fixtures as individual elements are now achievable, not just walls and slabs.
• Ask for open formats. IFC for records, USD if simulation or a digital twin is on the roadmap.

Moving Atoms is built on these changes: upload a scan from any scanner and get every object separated in Revit, IFC, STEP, USD and GLB within 48 hours, or open the LiDAR samples to see the results.

Frequently asked questions