IBM and NASA Open-Source a Lunar Foundation Model
Original: A rough guide for going back to the Moon
Why This Matters
Reliable lunar terrain intelligence is a prerequisite for any sustained human presence on the Moon or future Mars missions.
IBM and NASA released the NASA-IBM Lunar Foundation Model on September 10, 2026, an open-source multi-modal AI trained on decades of US and Japanese lunar mission data. It integrates sensors at scales ranging from 1 meter to 20 km per pixel to support crater mapping, volcanic history analysis, and polar ice prospecting for Artemis mission preparation.
As NASA's Artemis program works toward establishing a permanent lunar base—a stepping stone to Mars—accurate, comprehensive lunar mapping has become a practical necessity. The NASA-IBM Lunar Foundation Model is the first AI to unify observations across multiple data modalities, viewing angles, and spatial resolutions into a single reusable representation.
The model's architecture is based on TerraMind, previously developed by IBM and the European Space Agency for Earth observation. TerraMind was chosen for its ability to reconcile wildly different data types: NASA's GRAIL mission captured gravitational field data at 20 km-per-pixel resolution, while the Lunar Reconnaissance Orbiter imaged crater rims and boulders at 1 meter-per-pixel. Getting these datasets to 'talk to each other' is exactly what the model is designed to do.
NASA has identified three immediate priorities for the tool: cataloguing previously unmapped small craters; studying volcanic lava flow history; and locating ice deposits in permanently shadowed polar craters. That ice matters operationally—future crews could use it for drinking water, oxygen generation, and rocket fuel.
The Moon's lighting conditions add another layer of complexity. With two-week cycles of full sunlight and total darkness, and no atmosphere to diffuse light, surface features shift dramatically in appearance. The model is designed to handle these visual distortions, including those caused by regolith—the abrasive dust layer coating the surface.
All datasets and the model itself have been open-sourced. IBM Research Europe director Juan Bernabé-Moreno framed it plainly: 'We hope that our AI model can help the science community explore the lunar landscape and help the next generation of astronauts find their way around before heading into space.'