Real-Time lidar-IMU Odometry for Wheeled Robots via 3D-to-2D Feature Projection

Abstract

As a cutting-edge technology in the AGV industry, 3D lidar SLAM is widely applied in the navigation of wheeled robots due to its ability to provide precise and robust positioning for robots. Compared with 2D lidar SLAM, 3D SLAM often requires more computing resources. This paper proposes a novel method for extracting 3D features from point clouds based on non-repetitive scanning lidar. The 3D feature point clouds are projected into two dimensions, and the improved RANSAC algorithm is applied to a 2D point cloud, realizing robust 2D point cloud registration of unstable feature points. The algorithm was tested on forklift AGV platform. The results show that this method can extract feature points in both indoor and outdoor environments, providing odometry output, while consuming relatively limited computing resources. This research provides a novel approach for the implementation of real-time lidar-IMU odometry.