Globally Consistent and Tightly Coupled 3D LiDAR Inertial Mapping

TL;DR

This paper introduces a real-time 3D LiDAR-inertial mapping framework that tightly couples LiDAR and IMU data for accurate, robust, and efficient localization and mapping in challenging environments.

Contribution

It presents a novel framework combining GPU-accelerated voxelized GICP and IMU preintegration for global consistency and tight coupling in 3D mapping.

Findings

Achieves accurate localization in feature-less environments.

Demonstrates robustness on urban and college datasets.

Maintains low computational cost with fixed-lag smoothing.

Abstract

This paper presents a real-time 3D mapping framework based on global matching cost minimization and LiDAR-IMU tight coupling. The proposed framework comprises a preprocessing module and three estimation modules: odometry estimation, local mapping, and global mapping, which are all based on the tight coupling of the GPU-accelerated voxelized GICP matching cost factor and the IMU preintegration factor. The odometry estimation module employs a keyframe-based fixed-lag smoothing approach for efficient and low-drift trajectory estimation, with a bounded computation cost. The global mapping module constructs a factor graph that minimizes the global registration error over the entire map with the support of IMU constraints, ensuring robust optimization in feature-less environments. The evaluation results on the Newer College dataset and KAIST urban dataset show that the proposed framework enables accurate and robust localization and mapping in challenging environments.