Lunar Foundation Model
#8 en IA pour les sciences & la biologieNasa · depuis 10. September 2026 · 4× · vu le 13 sept. 2026
The NASA-IBM Lunar Foundation Model is an open AI foundation model for scientific exploration of the Moon, developed jointly by NASA and IBM Research together with several academic institutions. It is trained primarily on data from the Lunar Reconnaissance Orbiter (LRO) plus instruments from four missions in total (including LRO, GRAIL, and JAXA's SELENE/Kaguya), using roughly 2 million image tiles. The model is freely hosted on Hugging Face, its code is on GitHub, and it integrates with the open-source TerraTorch toolkit; it supports tasks such as crater mapping, ice prospectivity, and volcanic feature detection, and is part of the NASA-IBM AI4Science model family (alongside Prithvi and Surya).
Fonctionnalités
| Deployment Model | Self-hosted / open weights: download weights from Hugging Face, local fine-tuning via TerraTorch possible |
| Use Case Scope | Lunar science: crater mapping, ice prospectivity in shadowed craters, volcanic formation detection, supporting Artemis mission planning |
| Integrations | TerraTorch (fine-tuning toolkit), LoRA adapters for task checkpoints, Hugging Face collection with downstream models |
| License | Apache-2.0 (model weights); SomBench dataset under CC BY 4.0 |
| Platform | Hugging Face (model weights), GitHub (code), integrated with TerraTorch toolkit |
| Price | Free, openly available (open source) |
| Release Date | September 10, 2026 |