An Efficient Odometry and Mapping Framework Based on Multi-LiDAR Systems

Due to the accuracy and reliability of range measurements, LiDAR-based odometry and mapping have been the focus of research in autonomous driving. However, a single LiDAR is usually limited by the field of view (FoV) occlusion and data sparsity, which bring challenges for accurate pose estimation. To address this problem, we propose an efficient odometry and mapping framework based on a multi-LiDAR system. We first propose an online extrinsic calibration method that combines hand-eye calibration and optimization based on local map features. Then we adopt the sliding window with bundle adjustment to match secondary extracted structured features from fused point clouds for front-end odometry. In addition, we design a global manifold optimization strategy based on a factor graph with loop closure constraints and a weight noise model to optimize the states marginalized after the bundle adjustment. The experimental results indicate that our method achieves higher quality performance compared to other single LiDAR-based methods.