LiDAR Odometry for Autonomous Mining Trucks Utilizing Massif Registration

Minghan Zhou,Yafei Wang,Ruitao Zhang,Zexing Li,Yanran Luo,Xulei Liu

DOI: https://doi.org/10.1109/ICUS58632.2023.10318505

2023-01-01

Abstract:SLAM has been employed in the construction of unknown and complex environments in autonomous driving. However, special challenges exist in open-pit mining scenario: uneven terrains cause inaccurate sensing observation, especially LiDAR point cloud and IMU data; open-pit mines cannot provide valid geometric feature points and loop closure constraints, which leads to cumulative error. To address these problems, we present a LiDAR point cloud registration algorithm, including ground normal correction (GNC), 2D projection of massif alignment (PMA), and 3D massif alignment (MA), which calculates a more accurate pose in odometry. Our method exploits GNC to correct the potential deviation of roll and pitch angle by RANSAC, then PMA is performed to retrieve the 2D transform matrix. Finally, MA is introduced to implement 3D registration based on the prior pose from previous steps. We also present an extensive evaluation in real-world open-pit mines by constructing a SLAM front-end odometry, including the result of mapping and comparison of trajectories. Using GNSS data as ground truth, our proposed method shows higher accuracy on odometry without expensive time costs.