A Trade Study of Optical Global Localization Approaches for Lunar Rovers

Long-range planetary roving requires onboard, autonomous global localization to correct accumulated odometry drift without relying on ground-in-the-loop intervention. The Endurance lunar mission concept exemplifies these operational demands, requiring a solar-powered rover to traverse over 2700 km across the South Pole-Aitken basin on the lunar farside. High-resolution orbital imagery from the lunar reconnaissance orbiter serves as an opportunistic enabler for optical global localization, performing cross-view registration by correlating rover stereovision against geo-referenced orbital maps. However, the lunar surface presents severe perception challenges, such as low solar elevation angles, featureless terrain, and high-contrast shadows, that heavily degrade conventional vision algorithms. To evaluate optical global localization techniques for planetary rovers using the Endurance mission architecture as a reference, this trade study evaluates three candidate algorithms: crater-based localization (LunarNav and ShadowNav) and intensity-based registration (Censible). We introduce algorithmic enhancements adapting these frameworks to ingest snapshot 360-degree panoramic stereovision and execute during stationary phases. The adapted frameworks are systematically evaluated across 882 extensive scenarios within JPL's physics-driven Multi-fidelity Simulator for Endurance (MuSE), rigorously assessing positioning accuracy across varied terrain, stereo processing pipelines, and low sun elevation angles. This study establishes quantitative performance bounds for varied cross-view optical registration techniques, providing the technical justification required to enable autonomous global localization onboard planetary surface missions.