Robust and Accurate 3-D Point Cloud Registration Method for Close-Range Docking of Spacecraft

Abstract

Spacecraft docking at close-range is the final phase for contact between two spacecraft with nonnegligible relative motion, which requires precise pose estimation. Given the severe consequences that a docking failure can entail, the pose estimation algorithm for close-range spacecraft docking must achieve exceptionally high reliability. However, existing point cloud registration methods have difficulty accurately calculating transformations for data with outliers. We propose a method to estimate the pose of a spacecraft using a 3-D point cloud measured during close-range docking. The proposed method consists of region-of-interest (ROI) extraction and registration. In ROI extraction, the densities of feature points between the source and target point clouds are calculated, and candidate surfaces with high density are selected. Signature surfaces are then selected based on the geometric information of the candidate surfaces. In the registration process, rotation and translation are calculated using the selected surfaces, and the final transformation is obtained through a fine-tuning process. Specifically, this article proposes a rotation estimation method using the graduated nonconvexity optimization, which incorporates an adaptive parameter strategy to mitigate convergence to local minima during nonconvexity recovery. Experimental results verify that our method achieves higher accuracy and robustness to noise compared with other algorithms. In addition, we experimentally validated the algorithm using real alignment measurements obtained during the assembly, integration, and test phase of the Korea Pathfinder Lunar Orbiter.