BR$^{2}$O-LI: Bias Reduction and Bayesian Rotation- Speed Optimization for a LiDAR-Inertial System Toward Indoor Panoramic Reconstruction

Abstract

Mechanical light detection and ranging (LiDAR), which generate point clouds by spinning internal laser modules and timing their reflections, have gained significant interest from both academia and industry. However, their application in dense scanning scenarios, such as building information modeling and reconstruction, is limited by their narrow vertical field of view (FOV) and vertical resolution. Current methods increase the vertical FOV of LiDAR and improve resolution by rotating the LiDAR device. Although effective, they typically employ a fixed rotation speed based on prior experience, which may lead to suboptimal performance under varying environmental conditions during surveying. To address this challenge, we employ a simulation-assisted Bayesian optimization approach to dynamically adjust the rotation speed, ensuring superior point cloud density and uniformity within dynamic inspection settings. Moreover, we find that rotational movement inherently aids in the self-correction of IMU bias. Building on this, we develop a simultaneous localization and mapping (SLAM) framework that integrates a robust Kalman filter for high-rate motion tracking and a pose-graph optimization backend for prolonged stability. In addition, we introduce a novel rotation-modulated IMU preintegration factor, which effectively reduces IMU bias effects, further enhancing the accuracy and reliability of LiDAR-inertial SLAM for pose estimation and map construction. Comprehensive experiments are conducted in real-world settings to validate the effectiveness of the proposed method. In a dense panoramic mapping test of a <inline-formula><tex-math notation="LaTeX">$\text{5000}\,\text{m}^{2}$</tex-math></inline-formula> underground parking area, the system improves the spatial coverage, density, and uniformity of the point-cloud by <inline-formula><tex-math notation="LaTeX">$1.5\times$</tex-math></inline-formula>, <inline-formula><tex-math notation="LaTeX">$2.1\times$</tex-math></inline-formula>, and <inline-formula><tex-math notation="LaTeX">$1.6\times$</tex-math></inline-formula>, respectively, while achieving an absolute positioning error of <inline-formula><tex-math notation="LaTeX">$\text{0.37}\,\text{m}$</tex-math></inline-formula>.