LunarFM: A Shared Multimodal Representation of the Moon's Surface

Researchers introduced LunarFM, a multimodal foundation model for lunar remote sensing.

It learns a shared embedding from six orbital instruments across three lunar missions.

The team also released a co-registered dataset and pretrained embeddings for downstream lunar analysis.

LunarFM is designed to support tasks such as resource mapping, mineral abundance regression, and geological unit classification.