3D Keypoint Detectors from Scanning Mirror LIDAR Scans for Localization of Lunar Rovers*

This paper presents the implementation and evaluation of multiple 3D keypoint detectors on LiDAR scans for rover terrain relative navigation (TRN) on the Lunar surface. One method utilizes morphological dilation to detect positive objects while the other utilizes normals analysis and raytracing to detect craters. These detectors serve as front ends to many localization solutions by reducing a point cloud down to a set of salient keypoints that can be tracked between scans. A novel implementation of the Random Finite Set (RFS) based Rao-Blackwellized Particle Filter Simultaneous Localization and Mapping (SLAM) algorithm is used to process these keypoints due to the RFS framework's ability to robustly handle typical measurement characteristics statistics such as noise, missed detection, and false alarm with an elegant approach to the data association problem. More specifically, this paper implements the Probabilistic Hypothesis Density (PHD)-SLAM to Lunar rover TRN for two different 3D keypoint detectors. The proposed methods are evaluated using synthetic data obtained from a simulated micro-electromechanical system (MEMS) LiDAR in a procedurally generated Lunar environment based on the NVIDIA Omniverse frameworkImplementation details of this work is provided in github.com/TL-4319/RFS_SLAM_OmniLRS.