Gaussian mapping & scene graphs
LiDAR-enhanced 3D Gaussian reconstruction and structured scene representations, toward compact semantic maps for robot reasoning and interaction.
Lightweight embodied scene graphs →Embodied scene understanding & robotics
Huai Yu 余淮
Associate Professor & Ph.D. Supervisor · Wuhan University
We focus on how robots build compact, meaningful world representations to guide navigation, planning, control, and interaction.
I am with the Electronic Information School at Wuhan University. My work spans robot 3D perception, SLAM, navigation, and intelligent unmanned systems. My current research agenda explores lightweight embodied scene graphs that connect vision-language models with 3D Gaussian Splatting, toward task-oriented robot world models.
Previously, I was a Postdoctoral Fellow in AirLab at the Robotics Institute, Carnegie Mellon University, working with Prof. Sebastian Scherer. I am currently visiting Prof. Lihua Xie's group at the School of Electrical and Electronic Engineering, Nanyang Technological University (NTU), Singapore.
Research directions
From 3D perception to embodied understanding.
Multimodal perception & SLAM
Connecting cameras, LiDAR, and inertial sensing for calibration, localization, odometry, and geometric registration.
I2D-Loc, QuadricsReg & UniCalib →Event-based & robust vision
Geometric feature detection across camera models and challenging motion and illumination.
FE-LSD & ULSD →Current research agenda
Lightweight embodied scene graphs.
How can a robot move from seeing geometry to understanding a scene and acting on instructions? We aim to connect the open-vocabulary semantics of vision-language models with efficient 3D Gaussian representations.
- 01 / Represent
Semantic Gaussian maps
Explore incremental semantic–geometric mapping, with consistent open-vocabulary features and compact representations for resource-constrained robots.
- 02 / Understand
Task-oriented scene graphs
Abstract Gaussian maps into objects, spatial relationships, and functional attributes that support semantic queries and task reasoning.
- 03 / Interact
Language-guided action
Investigate bidirectional updates between symbolic scene graphs and 3D geometry for instruction-driven navigation, interaction, and scene editing.
A closer look
Selected work
Semantic 3D perception · CVPR 2026
Dr.Occ: Depth- and Region-Guided 3D Occupancy from Surround-View Cameras for Autonomous Driving
Depth and region guidance for semantic 3D occupancy prediction from surround-view cameras.
LiDAR-enhanced 3DGS · ICRA 2025
LiDAR-enhanced 3D Gaussian Splatting Mapping
Integrating LiDAR and images for Gaussian splatting mapping with geometric and photometric supervision.
Structured 3D geometry · TRO 2026
QuadricsReg: Large-Scale Point Cloud Registration using Semantic Quadric Primitives
Semantic quadric primitives for compact, large-scale point cloud registration.
Gaussian scene graphs · RAL 2026
SGE-GLoc: Semantic Gaussian Ellipsoid Scene Graphs for Efficient LiDAR Global Localization
Semantic Gaussian ellipsoid scene graphs for efficient LiDAR global localization.
Cross-modal localization · RAL 2025
I2D-LocX: An Efficient, Precise and Robust Method for Camera Localization in LiDAR Maps
Precise and robust camera localization using prior LiDAR maps.
Event-based vision · TPAMI 2024
Detecting Line Segments in Motion-blurred Images with Events
Combining events and images to detect line segments under motion blur.
Updates
Recent news
- 2026/09Our IROS 2026 papers explore lightweight LiDAR-enhanced Gaussian splatting and GLoCal sensor calibration.
- 2026/07
- 2026/02Our work "QuadricsReg: Large-Scale Point Cloud Registration using Semantic Quadric Primitives" was accepted by IEEE TRO.
- 2026/02Our work "Dr.Occ: Depth- and Region-Guided 3D Occupancy from Surround-View Cameras for Autonomous Driving" was accepted by CVPR 2026.
- 2025/12Our work "SGE-GLoc: Semantic Gaussian Ellipsoid Scene Graphs for Efficient LiDAR Global Localization" was accepted by IEEE RAL.
- 2025/11Our work "UniCalib: Targetless LiDAR-Camera Calibration via Probabilistic Flow on Unified Depth Representations" was accepted by WACV 2026.
Earlier updates
- 2025/09Our work "Robust Visual Odometry using Rigidly-Bundled Arbitrarily-Arranged Multi-Cameras" was accepted by IEEE RAL.
- 2025/06Our work "I2D-LocX: An Efficient, Precise and Robust Method for Camera Localization in LiDAR Maps" was accepted by IEEE RAL.
- 2025/01Our work LiDAR-enhanced 3D Gaussian Splatting Mapping (LiGSM) was accepted by ICRA 2025.
- 2024/12We released MCVO, a multi-camera visual odometry method on arbitrarily arranged cameras.
- 2024/08
- 2024/08
- 2023/11Our Work FE-LSD was accepted by IEEE TPAMI, thanks all the co-authors.
- 2023/06One paper of "PyPose: A Library for Robot Learning with Physics-based Optimization" was accepted by CVPR 2023.
- 2022/10One paper of "I2D-Loc: Camera Localization via Image to LiDAR Depth Flow" was accepted by ISPRS J P&RS.
- 2022/08One paper of "Targetless extrinsic calibration between stereo, thermal and laser sensors" was accepted by IEEE TIM.
- 2022/08Our paper "ULSD: Unified line segment detection across pinhole, fisheye, and spherical cameras" won The U. V. Helava Award, the best paper of ISPRS J. P&RS 2021.
- 2022/02Our work "Unified Representation of Geometric Primitives for Graph-SLAM Optimization Using Decomposed Quadrics" was accepted by ICRA 2022.
- 2021/06Our work "ULSD: Unified Line Segment Detection across Pinhole, Fisheye, and Spherical Cameras" got accepted by ISPRS J P&RS.
Publications
Research publications
Selected publications, ordered by year. Huai Yu is highlighted. * Corresponding author; # equal contribution.
2026
Lightweight Large-Scale 3D Reconstruction and Rendering using LiDAR-Enhanced Gaussian Splatting
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2026.
GLoCal: Global-Local Flow Estimation for Robust LiDAR-Camera Calibration
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2026.
AC-LIO: Toward Asymptotic Compensation for Distortion in LiDAR-Inertial Odometry via Selective Intraframe Smoothing
IEEE/ASME Transactions on Mechatronics, 2026.
QuadricsReg: Large-Scale Point Cloud Registration using Semantic Quadric Primitives
IEEE Transactions on Robotics, 2026
2025
Unsupervised Multiview UAV Image Geolocalization via Iterative Rendering
IEEE Transactions on Geoscience and Remote Sensing, 63, 2025.
2024
Detecting Line Segments in Motion-blurred Images with Events
IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2024
2022
I2D-Loc: Camera Localization via Image to LiDAR Depth Flow
ISPRS Journal of Photogrammetry and Remote Sensing, 2022.
2021
ULSD: Unified Line Segment Detection across Pinhole, Fisheye, and Spherical Cameras
ISPRS Journal of Photogrammetry and Remote Sensing, 2021. The U. V. Helava Award – Best Paper 2021
2020
Monocular Camera Localization in Prior LiDAR Maps with 2D-3D Line Correspondences
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2020.
Background
Education
- 2018–2020
Carnegie Mellon University
Visiting Ph.D. in Robotics
- 2015–2020
Wuhan University
Ph.D. in Communication and Information System
- 2011–2015
Wuhan University
B.S. in Electronic Information Engineering
Appointments
- Aug. 2026–present
Visiting Scholar
Nanyang Technological University
School of Electrical and Electronic Engineering - Sep. 2025–present
Associate Professor
Wuhan University · Electronic Information School
- Nov. 2021–Aug. 2025
Associate Research Fellow
Wuhan University · Electronic Information School
- Nov. 2020–Oct. 2021
Postdoctoral Fellow
Carnegie Mellon University · Robotics Institute
Academic service
Editorial & professional service
- Associate Editor, IEEE Robotics and Automation Letters (since Aug. 2025)
- Associate Editor, IROS 2023–2026
- Young Editorial Board Member, Robot Learning
- Young Editorial Board Member, Chinese Journal of Ship Research
- Area Chair, ICRA 2025 Workshop FMNS
- Session Chair, ICRA 2024
Reviewing & teaching
- Reviewer: IEEE TPAMI, TRO, RA-L, TIM, TGRS, IJRR; CVPR, ECCV, ICCV, ICRA, and IROS.
- Teaching: Introduction to Artificial Intelligence.
- Member: IEEE, CCF, and CSIG.
Honors & awards
ISPRS U.V. Helava Award, Best paper of ISPRS J. P&RS, 2021
ULSD publication →
More honors & awards
- Outstanding Student Advisor, National Graduate Electronic Design Contest, 2023.
- Supported by the Hubei Chutian Scholar Program and Wuhan Talent Program.
- National Scholarship, 2018
- Yugang-Songxiao Fellowship, 2018
- 1st Prize & Best Paper of National Graduate Electronic Design Contest, 2018
- Finalists & 6th Place of ICRA 2018 DJI RoboMaster AI Challenge, 2018
- Geospatial Technology Collaborative Innovation Scholarships, 2017
- 2nd Place of Breakthrough Competition in National Smart UAV IntelliSensing Technical Competition, 2017
- 3rd Place of Comprehensive Competition in National Smart UAV IntelliSensing Technical Competition, 2017
- Finalists & 6th Place of ICRA2017 DJI RoboMaster Mobile Manipulation Challenge, 2017
- National Scholarship, 2016
- Geospatial Technology Collaborative Innovation Scholarships, 2016
- Finalists & 3rd Place of 2016 DJI Developer Challenge
- The first-class prize of the national college students "Freescale cup" smart car competition, 2014
Join us & get in touch
Research opportunities
I supervise Ph.D. and master's students in robotics and intelligent unmanned systems. Prospective students and postdoctoral researchers interested in 3D perception, Gaussian mapping, scene graphs, and SLAM are welcome to contact me.
Graduate recruitment information (中文) ↗yuhuai@whu.edu.cn ↗
Electronic Information School · Wuhan University
Yugang-Songxiao Building
Wuhan, Hubei, China