Multi-sensor localisation and datasets
UWB-aided estimation and shared datasets for aerial and ground robot perception.
My work at NTU examined how cameras, inertial measurements and ultra-wideband (UWB) ranges can be combined for localisation. A recurring question is when sensor geometry makes an estimate reliable, and when it leaves important degrees of freedom poorly constrained.
UWB-aided estimation
The relative transformation paper studies robot-to-robot frame alignment from odometry and UWB ranges. It analyses a four-degree-of-freedom problem using the Fisher information matrix and Cramér–Rao bound. QCQP and semidefinite-relaxation solutions are evaluated in simulation and aerial-robot experiments. The author preprint is openly available.
VR-SLAM extends this research to a monocular camera and UWB anchors. It combines global alignment with range-aided bundle adjustment and pose-graph optimisation. The linked record is an arXiv preprint.
Related first-author papers cover odometry-scale and anchor estimation (2022), collaborative localisation (2022), camera–IMU–UWB fusion (2021), degenerate anchor configurations (2020) and single-anchor visual odometry (2020).
Shared research datasets
I co-authored the following dataset papers. Each official project page provides the data, calibration or evaluation resources and licence terms.
MCD · CVPR 2024
MCD covers campuses in Asia and Europe across different seasons. It includes spinning and non-repetitive LiDAR, stereo cameras, IMUs and UWB sensing. Survey-grade prior maps, continuous-time ground truth and semantic labels support evaluation of robot perception methods.
Publisher · Dataset and documentation
MMAUD · ICRA 2024
MMAUD supports drone detection, classification and trajectory estimation with visual, LiDAR, radar and audio measurements. The official repository documents the released recordings and sensor configuration.
Publisher · Dataset and documentation · Preprint
NTU VIRAL · IJRR 2022
NTU VIRAL records indoor and outdoor UAV flights using two 3D LiDARs, two synchronised cameras, IMUs and UWB ranging. The project provides ground truth and evaluation tools. Its documentation explains the offset between the IMU and the tracking prism, which matters when comparing trajectories.