Abstract
– For deep-space missions operating at large distances from Earth, autonomous relative optical navigation is emerging as a key technique because communication delays and the need for low-latency, high-precision control make ground intervention impractical. The Japan Aerospace Exploration Agency's (JAXA) Martian Moons eXploration (MMX) mission, scheduled for launch in 2026, depends on this capability to enable safe close-proximity operations around Phobos. Existing Phobos imagery, however, does not cover the viewing geometries and illumination conditions of the MMX close-proximity phase, limiting the construction of realistic datasets for vision-based navigation, guidance, and control. To bridge this gap, this work introduces an open-source, mission-agnostic optimisation framework built on top of the Blender-based Celestial Object Rendering TOol (CORTO), which automatically ingests PDS imagery, reconstructs the corresponding observation geometry, denoises the references, aligns synthetic and real frames in two passes via per-object identity masks, and tunes the shading parameters of each rendered body against a multi-metric composite cost function minimised by a Covariance Matrix Adaptation Evolution Strategy. Trained on six representative PDS frames and evaluated on a held-out test set of fourteen frames covering nine Phobos views from six distinct HRSC orbits and five Mars views, the converged parameters yield Phobos SSIM in the 0.78–0.93 range with a mean of approximately 0.85 and wide-context Mars SSIM in the 0.79–0.91 range, demonstrating consistent generalisation across viewing geometries. The optimised parameters are directly applicable to labelled synthetic datasets along reference Quasi-Satellite Orbits around Phobos.