Reimaging the Built Environment with Robotics and Artificial Intelligence

Team Members

RAIS4BE Lab at National University of Singapore pioneers robotic scanning and building information modelling (BIM) technology. Since the Lab started in 2021, we focused on LiDAR, image-based sensing, and semantic AI algorithms deployed on mobile robotic platforms for autonomous 3D scanning of legacy buildings or infrastructures, followed by high-precision 3D BIM reconstruction and semantic enrichment to develop content-rich engineering models for generative design and structural optimisation. Our lab brings together researchers from Civil, Mechanical, Architecture, Geodesy, Electrical and Computer Engineering.

Difeng (PhD/Postdoc)

Qiao Zheng

PhD Researcher

Kexin Li

PhD Researcher

Josh Li

PhD Researcher

Srivatsan

PhD Researcher

Yushuo Wang

MSc (Research)

Runfeng Ma

MSc (Research)

Dian Zhuang

Visiting PhD

Chao Xiang

Visiting PhD

Ruoming Zhai

Visiting PhD

Ben Ben

Lab Mascot

Mike Li

Research Fellow

Vincent Gan

Assistant Professor

Asiri

Research Associate @HKUST

Shaobo Li

Research Associate

Yuanyuan Deng

Research Assistant

Jey Chandar

Research Assistant

Tao Wang

PhD Researcher

Difeng Hu

PhD Researcher

Melanie Tan

PhD Researcher

Xiuqi Li

PhD Researcher

Oh Hui Lin

MSc (Research)

Xiayi Chen

MSc (Research)

Robot Dog & LiDAR Scanning (2022)

Automated 3D Mapping & Semantic Navigation

Trajectory Optimisation & 3D Perception (2024)

Our Research & Teaching

CDE Innovation Day Award

Teaching Excellent Award

Featured Publication

Deng, Y.Y., Gan, V.J.L.,* • Advanced Engineering Informatics (4.2026)

Motion-prior and Confidence-aware Gaussian Splatting (MCGS) SLAM for 3D scene reconstruction of indoor built environments

This paper proposes a motion-prior and confidence-aware Gaussian Splatting (MCGS) SLAM, which hardnesses a probabilistic motion-prior framework, confidence estimation mechanism, and adaptive keyframe selection to guide photorealistic 3D scene reconstruction.

Zheng, Q., Gan, V.J.L.,* Li, M.K., • Automation in Construction (1.2026)

Semantic instance segmentation and automated 3D BIM reconstruction for viaduct using LiDAR point clouds and weakly-supervised learning

This paper presents an AI-based semantic instance segmentation approach that leverages weakly-supervised learning for high-precision segmentation and automated BIM reconstruction of transport infrastructure, focusing on viaducts.

Li, M.K., Gan, V.J.L.,* Wang, B.Y., • Automation in Construction (11.2025)

Integrating hierarchical segmentation and vision-language reasoning for spatially complex and occluded MEP point clouds

This paper proposes a hierarchical and progressive segmentation that integrates deep learning-based semantic segmentation, geometry-driven instance segmentation, and vision-language model-assisted refinement for 3D BIM reconstruction of MEP systems.

Hu, D.F., Gan, V.J.L.,* • Automation in Construction (3.2025)

Semantic navigation for automated robotic inspection and indoor environment quality monitoring

This paper proposes a semantic navigation approach to improve robotic inspection. A revised RandLA-Net and KNN algorithm construct a semantic map rich in detailed object information. An object instance reasoning algorithm identifies and extracts target object coordinates from the semantic map. A semantics-aware A* algorithm calculates safer, efficient navigation paths.

Gan, V.J.L., Hu, D.F.,* etc. • Computer-Aided Civil and Infrastructure Engineering (3.2025)

Automated indoor 3D scene reconstruction with decoupled mapping using quadruped robot and LiDAR sensor

This study introduces an optimization algorithm incorporating viewpoint generation, occlusion detection and culling, and robot-moving trajectory identification. The research investigates 3D reconstruction, comparing coupled and decoupled approaches to identify most practical configuration for robotic scanning.

Gan, V.J.L., Li, K.X.,* etc. • Applied Energy (1.2025)

3D reconstruction of BIM with weakly-supervised learning for carbon emission modelling in the built environment

This paper presents weakly-supervised learning for automated BIM reconstruction, aiming at accurate carbon performance evaluation. By employing weakly-supervised semantic segmentation, this approach segments structural components from 3D point clouds and formulates the topological relationships of objects for BIM reconstruction to assess embodied carbon.

Zhai, R., Zou, J., Gan, V.J.L.,* etc. • Automation in Construction (10.2024)

Semantic enrichment of BIM with IndoorGML for quadruped robot navigation and automated 3D scanning

In this paper, BIM data schema is enriched with IndoorGML, integrating building geometry with spatial data to establish an indoor navigation model describing multi-scale spatial topological networks. This navigation model optimizes robot scanning positions and traversal sequences.

Wang, T., Gan, V.J.L.,* • Automation in Construction (10.2024)

Enhancing 3D reconstruction of textureless indoor scenes with IndoReal multi-view stereo

This paper presents the “IndoReal-MVS” dataset, a rich indoor-centric compilation reflecting real-world phenomena through advanced computer graphics. It introduces unsupervised “IndoorMatchNet”, synergising Feature Pyramid Network (FPN) and Pyramid Flowformer (PFF) for encoding complex indoor geometries.

Hu, D., Gan, V.J.L.,* etc. • Building and Environment (8.2022)

Multi-agent robotic system (MARS) for UAV-UGV path planning and automatic sensory data collection in cluttered environments

This paper presents a multi-agent robotic system for automatic UAV-UGV path planning and indoor navigation to automate sensory data collection. An enhanced shunting short-term memory model is proposed to optimise the pathfinding, 2D image and 3D point cloud data collection.

Gan, V.J.L.,* • Automation in Construction (2.2022)

BIM-based graph data model for automatic generative design of modular buildings

This paper presents a Building Information Modelling (BIM)-based graph data model for the theoretic representation of spatial attributes, topological relationships, geometries, and semantics for generative design of modular buildings.

Research Innovation

Forging New Frontiers - Robotic Scanning (cde.nus.edu.sg/cde-research-jan2025)

This research focuses on robot-assisted mobile scanning integrating upfront sensor perception with trajectory optimisation, semantic 3D mapping and navigation, and multi-view stereo / GS-SLAM for autonomous data acquisition in built environments.

  • Sensor perception and environment-aware trajectory optimisation

  • Semantic navigation and automated LiDAR mapping

  • Multi-view stereo and GS-SLAM for weakly-textured 3D reconstruction

  • Supported by CDE, our research has expanded into robot social navigation and geospatial intelligence, in collaboration with Civil, Electrical Engineering and FASS Geography. We are also pursuing a start-up on Scan2BIM Robot

This research aims to develop representation learning pipelines to transform 3D scan data into structured, semantically-enriched BIM models aligned with building foundation model requirements.

  • Representation learning and fine-grained semantic instance segmentation

  • Knowledge based inferencing for BIM reconstruction

  • Semantic similarity matching and relationship classification for semantic enrichment of BIM data

  • Digitalisation of transport infrastructure (supported by LTA), semantic-rich 3D model for lifecycle urban evaluation (supported by PUB), and ACMV energy anomaly detection and predictive control

Our computational design research focuses on graph-based representation and broad pre-training of 3D BIM geometry to capture structural priors and generate topologically valid designs capable of optimising the building structural performance.

  • BIM-based graph data model

  • Surrogate model predicting structural and environmental behaviours

  • Generative design methods to optimise shape & structural topology

  • BIM product library for digital fabrication and construction automation

  • Collaborator: Arup, HKCIC, Swire, NTU