Effect of LiDAR Mounting Parameters and Speed on HDL Graph SLAM-Based 3D Mapping for Autonomous Vehicles

Abstract

Generating high-accuracy 3D maps using Light Detection and Ranging (LiDAR) technology remains a critical challenge in autonomous vehicle (AV) development. While 3D mapping is foundational for reliable AV navigation, its accuracy is often compromised by poor LiDAR sensor calibration and external factors such as motion distortion. This study investigates the physical calibration of a LiDAR sensor mounted on a moving vehicle and its effect on 3D map generation using the HDL Graph SLAM algorithm. HDL Graph SLAM was selected as the offline post-processing method due to its self-correcting functions for estimating and auto-correcting positional errors from LiDAR data. Tests were conducted by varying the sensor tilt angle at -5°, 0°, +5°, and +10° and driving speeds at 20 km/h, 30 km/h, and 40 km/h. Results showed that a 0° angle at 30 km/h produced the most accurate 3D map, achieving a Root Mean Square Error (RMSE) of 0.0812 for straight paths and 0.1345 for curved paths. These findings demonstrate the significance of physical mounting parameters and speed on mapping performance. The study provides practical recommendations for LiDAR installation to enhance 3D mapping reliability under real-world road condition.