Feature Assessment and Enhanced Vertical Constraint Lidar Odometry and Mapping on Quadruped Robot

Jiehao Li, Dezhao Zeng, Qunfei Luo, Xiwen Luo, C. L. Philip Chen, Chenguang Yang

Abstract: Lidar simultaneous localization and mapping (SLAM) technology is a crucial cornerstone for intelligent robotics and autonomous driving applications. Despite significant advancements in lidar-based SLAM, traditional approaches suffer from notable error accumulation in the vertical direction. We propose a novel lidar odometry and mapping method incorporating feature assessment and reinforced vertical constraints to address this issue. In the lidar odometry framework, we introduce residual normal vectors to enhance vertical constraint enforcement and assign higher weights to stable feature points, thereby improving localization precision in the vertical domain. Furthermore, during local map optimization, we integrate the local map into an incremental voxel representation, traversing each voxel to evaluate the reliability of feature points. By discarding unstable features and amplifying the weights of stable ones, we elevate the coherence of the local map, providing subsequent lidar odometry with a more precise local map for optimal matching, Finally, we conducted experimental tests on the publicly available S3E dataset, as well as our own recorded datasets encompassing outdoor road scenarios, straight corridors, and underground parking facilities. The results demonstrate that, in comparison to existing SLAM methodologies, our proposed method achieves superior accuracy while maintaining a comparable computational load, thus significantly reducing vertical error magnitudes.