A Study on Efficient LiDAR-based Localization with Initial Pose Estimation

In this paper, we propose a method of obtaining an initial pose estimate in a simple and computationally efficient manner for LiDAR scan matching such as ICP and NDT. When two scans are obtained at largely different two LiDAR poses, numerical optimization for scan matching tends to fall into a local minimum, and consequently we often suffer from incorrect matching. By using the proposed method, we can start to search for the relative pose between two scans with a sufficiently accurate initial pose estimate, and obtain correct matching through numerical optimization. The proposed method is derived theoretically under some assumptions, and its effectiveness in actual indoor environments is examined by experiments.