NASA, IBM launch new AI model for studying the moon

NASA and IBM Launch Open-Source Lunar Foundation Model

Called the NASA-IBM Lunar Foundation Model, the system is hosted publicly on Hugging Face for free use, while its complete codebase is available on GitHub for testing and experimentation. The project forms part of an ongoing collaboration between NASA, IBM Research, and several academic institutions to bring artificial intelligence to planetary science.

Training Data and Core Capabilities

Unlike traditional models that require building and training specialized algorithms from scratch for individual assignments, foundation models are pre-trained on vast, unlabeled datasets. This broad pre-training allows them to generalize across multiple scientific domains through quick fine-tuning.

According to Space, the model was primarily trained on data collected over 17 years by NASA’s Lunar Reconnaissance Orbiter (LRO). The LRO mission has compiled a nearly complete, high-resolution mosaic of the lunar surface, generating data that exceeds all of NASA’s other planetary missions combined. Additional data sources include more than 30 layers of information gathered by nine instruments across four missions, consolidating decades of observations from U.S. and Japanese flights.

The foundation model integrates observations captured in a range of modalities, viewing angles, and spatial scales. In benchmark tests, it identified key features on the lunar surface up to 23 percent more accurately than widely used methods, according to Reuters.

Key Research and Exploration Priorities

NASA has initially prioritized three main tasks for the lunar AI model:

• Mapping smaller, uncatalogued craters to help select safe landing sites
• Investigating the moon’s volcanic history and identifying young volcanic features
• Scouring craters at both lunar poles to estimate the stability and location of water ice

Lunar ice is of significant interest to space agencies because it can provide future astronaut crews with drinking water, breathable oxygen, and rocket fuel for onward missions to Mars. The model can help scientists estimate where ice patches are likely to remain stable on and below the surface, even in challenging polar environments where extreme temperature shifts and lack of atmosphere create visual distortions.

Context and Future Lunar Exploration

The launch addresses a longstanding challenge for scientists handling high-volume, multi-sensor data at varying scales.

NASA has spent decades building an extraordinary scientific record of the moon, but collecting data is only part of the job, Kevin Murphy, NASA’s chief science data officer and acting chief data and AI officer, said in the announcement. The NASA-IBM Lunar Foundation Model shows what's possible when we bring AI to NASA's petabytes of scientific data. That's a real opportunity we see with AI: turning large-scale data into new discoveries.

The tool joins IBM and NASA’s Prithvi family of open foundation models, which already includes applications focused on Earth observation and heliophysics. Juan Bernabé-Moreno, director of IBM Research Europe for Ireland and UK, noted that the model aims to help the science community explore the landscape and assist future crews in navigating the terrain.

NASA’s Artemis program plans to return astronauts to the moon, testing new technologies to build a base where crews can live and work, conduct research, and prepare for future exploration.