Lunar-FM World’s first AI foundation model built specifically for the Moon
The Frontier Development Lab (FDL.ai) LunarLab, a partnership between the Luxembourg Space Agency, the European Space Resources Innovation Centre (ESRIC), and Trillium Technologies, announced Lunar-FM, the first AI foundation model dedicated to lunar exploration and resource prospecting.
Developed with advanced computing and technical support from Google Cloud, NVIDIA, SCAN Computers International Ltd, and Datarock, Lunar-FM represents a major leap forward in how scientists analyse the Moon’s complex and fragmented data landscape
By applying state-of-the-art AI techniques, similar to those powering large language-based models like ChatGPT, Lunar-FM integrates vast and diverse lunar datasets into a unified, intelligent framework, known as a ‘foundation model’. These new class of science assets can be adapted to many downstream use-cases with the goal to accelerate scientific discovery and in-situ resource utilization (ISRU) on the Moon.
Lunar-FM solves the Lunar Data ‘fragmentation problem’
For decades, lunar science has relied on data scattered across multiple missions and instruments, from optical imagery to gravity measurements. This fragmentation has made large-scale analysis slow and labor-intensive.
Lunar-FM addresses this challenge by combining 18 distinct data layers ingesting a vast array of diverse data sources; optical imagery from the Lunar Reconnaissance Orbiter (LRO), topography from LOLA, thermal data from Diviner, radar reflectivity from Mini-RF, and gravity anomalies, creating a unified multimodal representation of the Moon’s surface up to 70° north and south latitude.
Lunar-FM compresses these inputs into a powerful 768-dimensional embedding that captures geological, thermal, text and compositional properties in a unified way for the first time.
Breakthrough Capabilities for Lunar Science and Resource Exploration
Lunar-FM has already demonstrated several groundbreaking capabilities:
Mapping the Moon’s Titanium
The model can generate global predictive maps from extremely limited ground-truth samples. In one validation study, Lunar-FM mapped Titanium Dioxide (TiO₂) abundance across the Moon using just eight labeled Apollo samples.
Mapping the Moon’s Titanium
The model can generate global predictive maps from extremely limited ground-truth samples. In one validation study, Lunar-FM mapped Titanium Dioxide (TiO₂) abundance across the Moon using just eight labeled Apollo samples.
Filling Sensor Gaps and Enabling Advanced Search
By learning correlations between data types, the model reconstructs missing measurements and enables ‘similarity searches’ (the same technique you use when you search with an image on Google.) This helps scientists instantly identify regions with comparable geological features.
Geological Boundary Definition
Lunar-FMs data driven representations align closely with the USGS Unified Geologic Map of the Moon, demonstrating its ability to encode meaningful compositional and structural information.
Compressing vast data volumes that can be used on a PC
Lunar-FM achieves up to 300x compression, transforming massive multi-instrument datasets into lightweight embeddings that can be analyzed on standard computers.
“Talking to the Moon”: The Lunar Agentic Analyst
To make this powerful technology accessible, Lunar-FM includes a Lunar Agentic Analyst, a natural language interface that allows users to interact directly with lunar data.
Researchers and mission planners can now ask questions such as:
“Identify regions geologically similar to the Apollo 11 landing site.”
The system routes these conversational queries to the appropriate analytical tools within the AI model, dramatically simplifying complex scientific workflows and supporting real-time mission planning and rover operations.
Bridging Science and Operations
Dr. Abigail Calzada Diaz, Lunar Geologist at ESRIC, highlighted the impact of the new platform:
“The fragmented, multi-source nature of lunar data has made investigations labor-intensive. Lunar-FM provides a standardized, unified data infrastructure enabling knowledge extraction and synthesis across disparate datasets. Lunar-FM is a translational bridge between scientific understanding of lunar processes and the operational requirements of missions — turning theory into resource-focused targets.”
A Commitment to Open Science
In line with FDL LunarLab’s open science principles, Lunar-FM’s pre-trained embeddings and analytical tools will be released publicly following peer evaluation.
The official public launch will take place during Luxembourg Space Resources Week, 4–7 May 2026.
Researchers can already request early access through https://lunarlab.ai/, where technical briefings and detailed results are also available.
A Major Step for Luxembourg’s Lunar Innovation Ecosystem
With Lunar-FM, Luxembourg strengthens its position at the forefront of space resource innovation, AI-driven planetary science, and sustainable exploration.
By unifying decades of lunar data into an intelligent, accessible platform, LSA and its partners are enabling faster discoveries, smarter missions, and a new era of data-driven exploration of the Moon.
Read the official Press Release here (Pdf, 3.61 Mb).
Last update