Zeng Qingcheng<br/>曾庆铖

Zeng Qingcheng
曾庆铖

Time to get some curiosity!

Publications

SCORE: Saturated Consensus Relocalization in Semantic Line Maps

Published in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2025), 2025

A robust visual relocalization framework that addresses the challenges of large outlier ratios and excessive map storage requirements using semantic line representations. The method employs Perspective-N-Lines with a saturated consensus approach and accelerated branch-and-bound algorithm for two-stage rotation and translation estimation, achieving robust performance even with 99% outlier matches.

Download here

Distributed Invariant Kalman Filter for Object-level Multi-robot Pose SLAM

Published in IEEE International Conference on Robotics and Automation (ICRA 2025), 2025

A distributed left-invariant Kalman filter approach that utilizes semantic object-level 6-DoF estimations from Pose-CNN outputs and relative pose measurements between robots as filter observations. The method addresses uncertainty estimation issues caused by unknown correlations in inter-robot pose estimation through covariance intersection, ensuring system robustness when partial robots experience observation degradation.

Download here

Consistent and Optimal Solution to Camera Motion Estimation

Published in IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2024

A novel approach for camera motion estimation that achieves asymptotic Gaussianity in noise distribution through special error construction for essential matrix estimation. The method enables bias elimination through variance estimation and achieves maximum likelihood optimal estimation via single-step Gauss-Newton iteration, reaching the Cramér-Rao lower bound.

Download here

Efficient Invariant Kalman Filter for Inertial-based Odometry with Large-sample Environmental Measurements

Published in IEEE Transactions on Robotics (TRO, under review), 2024

An invariant Kalman filter design that models error distribution on the SE₂(3) manifold rather than traditional SO(3)×ℝ³ or SE(3), making the dynamics in error states become linear autonomous systems. This improves observer consistency and eliminates linearization errors. The framework demonstrates superior robustness in sensor-degraded scenarios such as long corridors and white walls.

Download here