InfraTwin
Robotics and AI for Automated Digital Twinning of Buildings and Infrastructure

From physical infrastructure to engineering intelligence.

A connected workflow from robotic reality capture to AI-assisted digital twins — developed at NUS College of Design and Engineering.

InfraTwin overview: autonomous robots and sensors capturing infrastructure as physical hardware, layered into a cognitive digital twin with geometry, semantic and condition layers All Rights Reserved
The Problem

Where and how time and money are lost

Ageing infrastructure assets, climate hazards and limited AEC industry capacity are converging into one problem for smart cities. Building and infrastructure owners need to survey the site, capture complete and up-to-date as-built conditions, and convert the captured data into usable digital records. This process faces two key bottlenecks.

Tracked scanning robot capturing 3D geometric data underneath a concrete infrastructure structure AI-generated
1.1

Reality Capture

Capture 3D geometric data (such as point clouds) and updated as-built conditions (RGB / thermal imagery) for infrastructure.

3D point cloud viewer software showing a processed scan of an industrial facility AI-generated
1.2

Incomplete Coverage

Achieving complete, reliable scan coverage across complex infrastructure is highly time-demanding. Incomplete coverage means missing regions, unreliable records, and repeated site visits.

3D point cloud viewer software displaying a raw captured scan of an industrial building, awaiting conversion into a digital record AI-generated
2.1

Digital Twinning

Captured data (e.g. point clouds) is processed manually by specialised experts to convert it into usable digital records and engineering 3D models for maintenance planning.

A specialist manually reviewing and annotating point cloud and drone survey data across multiple monitors at a workstation AI-generated
2.2

Manual Intervention

Cleaning, annotating and converting survey data into usable digital records still requires substantial specialist effort — accuracy and processing efficiency of existing semi-automated tools remain limited.

Our Solution & Innovation

InfraTwin closes the loop

A connected workflow from robotic capture to AI-assisted digital twins.

Perceive Understand Deliver
InfraTwin's autonomous scanning robot All Rights Reserved
01

Infra.SENSE

Identifies insufficiently scanned regions early and guides specialists/robots to capture more complete infrastructure data.

Point cloud segmented into structural, M&E and infrastructure components All Rights Reserved
02

Infra.SEG

Automatically segments point clouds into structural, M&E and infrastructure components.

Pipeline from raw point cloud input, through AI processing, to reconstructed BIM output All Rights Reserved
03

Infra.DT

Reconstructs IFC-compatible engineering digital twins for asset management.

All Rights Reserved
Real robot footage — indoor 3D scanning across stairs and a GPS-denied corridor, not a render.
Innovation 1

Infra.SENSE: Robots that know where to scan

The robot is not only a mobile scanner; it reasons about visibility, occlusion and scan coverage.

In complex infrastructure, surfaces are occluded, corridors are narrow, and the best viewpoint isn't always obvious. InfraTwin's sensor perception model predicts visibility and scan coverage before the robot moves, then selects viewpoints and trajectories that improve the final point cloud.

40–50%
reduction in scanning time
10–20%
improvement in scan coverage

Commercial meaning: fewer return visits, fewer missing areas, and more complete infrastructure records.

Real segmentation results comparing ground truth and predicted point cloud classification, Connection Hall and Lecture Hall All Rights Reserved
Real segmentation output — ground truth vs. predicted, across two building interiors.
Innovation 2

Infra.SEG: Raw scans become engineering digital twins

The platform turns point clouds into segmented components and reconstructed BIM objects.

Infra.SEG uses BIM-generated synthetic point clouds, weak supervision and vision foundation models to reduce annotation effort — adapting segmentation across industrial facilities, bridges, transport assets and complex M&E systems, without depending on manually labelling every new site from scratch.

Isometric building reconstructions with numbered defect and crack-detection callouts, derived from segmented point clouds All Rights Reserved
Segmented components carried through to condition/defect detection on a real heritage building survey.
Innovation 3

Infra.DT: Segmented components become IFC-compatible digital twins

Infra.DT moves from raw scan data to object-level information, ready for engineering and asset-management workflows.

For transport infrastructure, the system identifies structural and M&E components — beams, columns, ducts, pipes, equipment — from LiDAR point clouds and reconstructs them into IFC-compatible BIM. The same underlying segmentation-to-reconstruction pipeline extends to condition and defect assessment, shown here on a real heritage building survey — carrying labelled components through to fine-grained surface and structural condition detection.

Watch the Demo

See InfraTwin in action

Meet the Team

College of Design and Engineering, National University of Singapore

Li Jingxuan (Joshua)
Li Jingxuan (Joshua)
PhD Student, NUS

PhD research in robotics and AI for automated infrastructure digital twinning — developing InfraTwin's capture-to-digital-twin pipeline, including 3D scene understanding and the NUS3D point-cloud dataset.

Master's research: Research Assistant, Solar Energy Research Institute of Singapore (SERIS), Mar 2024–Jan 2025 — BIPV integration onto heritage buildings, with NHB & NUS Baba House Museum.

Design training: MArch, National University of Singapore (2023–2025). BArch, Chongqing University (2020–2023).

Speaker, EEHB 2024 (5th International Conference on Energy Efficiency in Historic Buildings, ICOMOS Singapore). Presenter, 43rd ISARC / 34th IGLC, Singapore 2026.

Dr Vincent Gan
Vincent Gan, PhD
Assistant Professor, Dept. of the Built Environment, NUS
Affiliations:
  • Assistant Professor, National University of Singapore, Singapore (09.2020 – present)
  • Member, Institute of Electrical and Electronics Engineers (IEEE), US (10.2024 – present)
  • Member, American Society of Civil Engineers (ASCE), US (01.2019 – present)

Second Runner-Up, NUS CDE Impact Accelerator Challenge (2024) — Scan2BIM: robotic scanning for automated 3D digital reconstruction.