Abstract: LiDAR SLAM plays a vital role in the automation of engineering vehicles, yet most existing algorithms are designed for static environments and face challenges when applied to dynamic construction scenarios. To address this, this paper proposes a novel dynamic LiDAR SLAM algorithm, Dyna-LIO, inspired by amphibian vision principles. First, a bio-inspired mapping framework is established to systematically describe the theoretical connection between biological principles and SLAM modules. Then, linear interpolation between LiDAR odometry and IMU odometry is performed to initialize the pose of the current frame; ground points are extracted and the point cloud is transformed into depth images and bird’seye views for dynamic object detection and removal. Finally, geometric feature-based registration of static point clouds is carried out, followed by global optimization using a factor graph to obtain high-precision LiDAR odometry results. Validation on the KITTI dataset shows that the proposed method achieves significant improvements in localization accuracy compared with existing algorithms in dynamic environments. Simulation experiments in typical construction and mining scenarios also demonstrate a substantial reduction in localization errors, further verifying the applicability and robustness of Dyna-LIO in real operational environments.

Key words: Biomimetic')">Biomimetic, Engineering vehicle, LiDAR SLAM, Dynamic point recognition, Vehicle localization