NASA and IBM lunar AI exploration collaboration

NASA and IBM have collaborated to release an open-source artificial intelligence tool known as the NASA-IBM Lunar Foundation Model. Designed to accelerate lunar scientific exploration, the model utilizes data from multiple NASA missions and has shown up to 23% greater accuracy in identifying geographic features compared to previous methods.

The model was trained on 17 years of data from NASA’s Lunar Reconnaissance Orbiter (LRO), including 2 million image tiles, alongside data from the GRAIL mission, Lunar Prospector, and Japan’s Selenological and Engineering Explorer. In comparative tests, the model demonstrated a 23% reduction in errors compared to Microsoft’s SwinV2-B system and outperformed it by 19% in crater recognition using only half the training data. It also successfully identified a new impact crater resulting from a SpaceX Falcon 9 collision.

Following the release of the foundation model, the partnership has expanded its application to develop detailed 3D digital maps of the lunar surface. Using computer vision and automatic shadow recognition, this technology processes satellite imagery to identify terrain features such as craters, rocks, and slopes. These mapping capabilities are intended to facilitate autonomous spacecraft navigation and assist NASA’s Artemis program in locating critical resources, such as water ice deposits in shadowed regions of the lunar south pole. To support the scientific community, the codebase is available on GitHub, and the model is hosted on Hugging Face for public use.