NASA and IBM release open-source AI model for lunar research
The Lunar Foundation Model is designed to analyse lunar imagery and instrument data, with a co-registered dataset released for researchers.
NASA and IBM have released an open-source artificial intelligence model designed to analyse the Moon, giving researchers a tool to identify areas where ice may be present and classify lunar craters.
The NASA-IBM Lunar Foundation Model is available to download through Hugging Face. The organisations also released a co-registered dataset containing tens of thousands of images and instrument data from NASA’s Lunar Reconnaissance Orbiter and GRAIL missions, as well as Japan’s SELENE mission.
NASA and IBM said the model reduced ice-detection errors by 23 per cent compared with SwinV2-B, a vision system used as a baseline for image-analysis tasks. In crater classification, the model reportedly improved results by 19 per cent while using half the training data.
The team trained and tested the system on geographically separated sections of the lunar surface. The approach was intended to account for changing illumination and the similarity of craters viewed from orbit.
IBM said the dataset contains more than two million corresponding data points across different types of lunar information. The release is intended to support lunar science and exploration during the Artemis era, although the source material does not detail the dataset’s licence or access conditions.
