State-of-the-art lunar data

The most detailed lunar digital elevation model (DEM) is currently based on the Lunar Reconnaissance Orbiter (LRO), it represents the surface of the Lunar north and south pole areas with a resolution of approximately 5m per pixel using interpolation techniques. Maps generated from the laser altimeter (LOLA) data naturally lose their
resolution rapidly as moving away from the poles.

Higher resolution maps can be generated with more advanced sensor fusion techniques like Photogrammetry and Photoclinometry of the Narrow Angle Cameras (NACs) images. In this way, a resolution of up to 1m per pixel can be achieved. Ideally, not only topographical information but also an albedo map can be extracted to a limited extent.

Despite all this, there is not yet enough real image information available to train an artificial intelligence for pin-point automatic landing and safe avoidance of hazards (small craters, boulders).

Existing DEMs provide topographical data but lack complementary high-fidelity albedo information, reducing their effectiveness for optical navigation reliant on surface texture recognition.

Current lunar landers have largely avoided low-angle illumination landings (<10° Sun angle) due to these limitations, underscoring the need for an enhanced dataset pipeline capable of generating training material for
AI-based precision landing and terrain analysis.

This DLVS3 extension addresses these challenges by integrating real and synthetic datasets, ensuring higher accuracy, adaptability, and robustness for next-generation lunar exploration missions.