Range-based Scan Matching for LiDAR-SLAM with Non-repetitive Omnidirectional LiDAR Masafumi Nakagawa(a*), Kenshiro Yamamoto(a), Tetsu Yamaguchi(a), Nobuaki Kubo(b), Etsuro Shimizu(b)
a) Shibaura Institute of Technology, Japan
*mnaka[at]shibaura-it.ac.jp
b) Tokyo University of Marine Science and Technology, Japan
Abstract
There are two main types of omnidirectional 3D LiDAR. The first type is a horizontally scanning LiDAR that performs repetitive linear scans. The second type uses micro-electro mechanical system mirrors to produce non-repetitive, non-horizontal scan lines, enabling the acquisition of wide field-of-view point clouds. Although repetitive-scanning omnidirectional LiDAR can capture high resolution data across a wide horizontal range, it is limited by a narrow vertical field-of-view and low vertical angular resolution. To achieve uniform spatial resolution for LiDAR-SLAM applications, multiple LiDAR sensors are often be combined or the LiDAR is physically rotated. Conventional LiDAR-SLAM techniques typically rely on repetitive scanning LiDAR and apply scan matching algorithms, such as the iterative closest point methodology, to align sequential scan lines. These methodologies benefit from the presence of numerous candidate correspondences between adjacent scans and employ point-to-point, point-to-line, or point-to-plane matching strategies. In contrast, although it requires a longer acquisition time, a non-repetitive and omnidirectional LiDAR achieves high angular resolution in both the horizontal and vertical directions. Consequently, a non-repetitive and omnidirectional LiDAR can reduce the number of 3D-LiDAR units needed and omit the need for rotational mechanisms when capturing wide-area point clouds. However, conventional scan-matching techniques are less effective for SLAM with non-repetitive LiDARs because the adjacent scans have few overlaps. Existing approaches address this limitation by incorporating motion distortion correction using inertial measurement units for tightly coupled position and attitude estimation. These approaches also use scan matching via point-to-plane methods based on multiple surface normals. Additionally, scan-to-map alignment registers local scans with global 3D maps. This study investigates the use of a non-repetitive, omnidirectional LiDAR for SLAM applications to develop a more compact and efficient LiDAR-SLAM system. To enable robust SLAM processing in environments lacking planar surfaces, we propose an range-based point-to-point scan matching methodologt tailored to non-repetitive omnidirectional LiDAR. We verified the feasibility of the proposed methodology using data acquired from a non-repetitive omnidirectional LiDAR mounted on a boat.
