Lunar Terrain Relative Navigation: Integration of Vision-Based Methods
Hyunsung Kim, Sunwoo Mun, Seokju Lee
ํ๊ตญ์ฐ์ฃผ๊ณผํํ ํ์ ๋ํ(๊ตฌ๋๋ฐํ), Apr 2026
Terrain relative navigation (TRN) systems are essential for precise lunar landing in GPS-denied environments. However, standalone relative visual odometry suffers from inherent scale ambiguity and accumulated drift over extended durations, while map-referenced absolute localization remains highly sensitive to illumination changes and prior map uncertainties. To overcome these fundamental limitations, we propose an integrated TRN framework that dynamically fuses high-rate relative motion estimation from input image sequences with low-rate absolute localization based on the pre-built lunar crater catalog. By employing an extended Kalman filter (EKF), the proposed architecture continuously estimates the spacecraftโs trajectory while constraining the metric scale and suppressing translational drift. Furthermore, we introduce a QuickMap-based lunar landing dataset targeting the lunar south pole, comprising four distinct illumination conditions with varying solar azimuth and incidence angles. Experimental evaluations demonstrate the feasibility of resolving scale ambiguity and improved localization stability compared to conventional frame-to-frame or map-only baselines, demonstrating its suitability for vision-based autonomous landing in resource-constrained spacecraft systems.
Lunar Landing Site Evaluation: Self-Organizing Maps with Temporal Solar Visibility
Junseo Moon, Seokju Lee
ํ๊ตญ์ฐ์ฃผ๊ณผํํ ํ์ ๋ํ(๊ตฌ๋๋ฐํ), Apr 2026
We implement a lunar landing site evaluation pipeline based on a self-organizing map and extend the baseline formulation, which uses mean solar visibility, by incorporating temporal solar visibility information. In the baseline formulation, solar visibility is represented by a time-averaged mean value, which does not capture the temporal structure of illumination and darkness intervals. Following prior topography-based illumination studies, we use a digital elevation model to estimate terrain-induced horizon obstruction in selected local regions of the lunar south polar area and combine it with SPICE-based solar geometry to model solar visibility over one synodic month (approximately 29.5 days) at fixed time intervals. From the resulting binary visibility time series, we derive raster features including average sun visibility, the longest period with solar illumination, and the longest continuous period in total darkness. We then incorporate these features as additional inputs to the self-organizing map to generate updated suitability and ranking maps. We examine four local south polar regions, namely Haworth, Shackleton, Cabeus, and Shoemaker, selected with reference to previously proposed candidate landing areas in the literature. Comparisons between baseline and temporally extended analyses suggest that temporal solar descriptors may alter local suitability patterns when similar mean visibility corresponds to different darkness-duration profiles. Furthermore, scenario-specific hourly analysis reveals that static suitability does not guarantee persistent operational availability over a one-month illumination cycle, highlighting the importance of temporal illumination modeling for local-scale lunar landing site analysis.