Answers are grounded in this paper and cite their sources.
Autonomous spacecraft proximity operations — rendezvous, docking, on-orbit servicing, and debris removal — require vision-based relative pose estimation, yet most deep learning methods provide only deterministic predictions. In safety-critical scenarios, overconfident errors can lead to mission loss. Uncertainty-aware deep learning has therefore emerged as a key research direction. This paper presents a systematic review of uncertainty-aware deep learning for monocular spacecraft relative pose estimation. Following PRISMA guidelines, 159 publications (published mainly between 2015 and 2026, plus a small number of foundational aerospace references) were included and critically analysed. The review first establishes a taxonomy of aleatoric and epistemic uncertainty, then critically examines major quantification paradigms — Monte-Carlo dropout, deep ensembles, heteroscedastic regression, evidential deep learning — with respect to mathematical formulation, computational cost, calibration quality, and compatibility with flight hardware. Uncertainty-aware domain adaptation, safety-critical deployment considerations (out-of-distribution detection, fault detection and isolation, certification-aware design, and layered propagation of uncertainty to guidance, navigation, and control) are analysed. Benchmark datasets are systematically compared, and current evaluation practices are appraised, revealing persistent gaps in realism, sensor modelling, and calibration reporting. A comparative synthesis of representative approaches leads to a decision framework and design guidelines. Open challenges — dataset realism, onboard safety assurance, active uncertainty reduction, explainability, and certified domain adaptation — are identified. The review concludes that while uncertainty-aware deep learning shows considerable promise, substantial work remains in flight validation, calibration standardization, and formal certification before these methods can be trusted for fully autonomous safety-critical spacecraft operations.
No takes yet. Share an insight, caveat, or question.
Koushafar et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: