Point cloud completion based pose estimation for spacecraft

Abstract

The growing density of near-Earth space operations necessitates advanced autonomous Guidance, Navigation, and Control (GNC) systems capable of reliable and precise pose estimation during close-proximity operations (CPO). Existing 3D vision-based localization methods, including traditional algorithms and recent machine learning enhancements, often struggle with noisy, sparse, or incomplete point cloud data, limiting their applicability in safety-critical, non-cooperative scenarios. This paper introduces PCC-KIPE, a novel pose estimation framework that overcomes these limitations by combining 3D point cloud completion with a reward-based search algorithm, eliminating the need for exact point correspondences or covariance matrices required by conventional methods. By reconstructing complete target models from partial observations and performing pose estimation through a robust search strategy, PCC-KIPE demonstrates enhanced resilience to data imperfections and reduced reliance on extensive training datasets. The proposed approach offers a promising direction for scalable, autonomous GNC in increasingly complex space environments.