The rapid growth of the space market is introducing many technical challenges in ensuring the long-term sustainability of the orbital environment and mission safety. Many solutions are being investigated, ranging from on-orbit servicing to active space debris removal, increasing the need for reliable and robust relative navigation systems. In particular, accurate monocular 6-DoF pose estimation of spacecraft represents a key enabling capability for future proximity operations. This thesis investigates learning-based monocular vision approaches for spacecraft pose estimation under realistic illumination conditions and domain shifts. The study is conducted using the SPEED+ benchmark dataset, which combines large-scale synthetic imagery with Hardware-in-the-Loop (HIL) acquisitions, enabling systematic evaluation across synthetic-to-real transitions.