NASA and IBM Launch Lunar AI Model and New Dataset

Key Takeaways

1IBM and NASA release an open-source AI model and a lunar dataset presented as new on Hugging Face

2The model, trained on decades of observations, claims up to 23% better performance in identifying major geographical features

3The initiative aims to provide a common foundation for the scientific community for research related to Artemis

Why it matters : These open-source tools are intended to accelerate and facilitate lunar exploration by enabling the scientific community to analyze data collected during the Artemis missions more efficiently.

An AI model trained on decades of lunar observations and a dataset presented as unprecedented are now available as open-source. Hosted on Hugging Face, they are intended for exploration within the Artemis program, with a claimed performance up to 23% better at mapping key terrains and identifying areas potentially rich in ice.

An open dataset and ice detection at the heart of the project

Researchers from IBM and NASA report the availability of what they consider the first open-source lunar dataset of this kind, comprising several tens of thousands of maps and images. This information results from data collected by nine different instruments during four distinct lunar missions. The associated model leverages multiple sources of observations to identify areas likely to contain ice. According to the press release, it outperforms widely used methods for identifying major geographical features, such as craters, volcanic formations, or potential ice deposits, by up to 23%. These elements are presented as essential for future Artemis explorations.

Less manual sorting, more large-scale patterns detected

Until now, researchers had to manually sift through maps and images from various instruments or rely on low-resolution machine learning models limited to a single task. The new model is touted as a time-saver by detecting large-scale patterns beyond a single instrument. It is also proposed as a foundation for building new research, rather than developing a separate AI model for each study.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Open-source access on Hugging Face and stated goals for Artemis

The NASA-IBM Lunar Foundation model is available as open-source on Hugging Face, following a launch that took place on a Thursday. It has been trained on decades of observations selected by researchers from both organizations. It is presented as one of the first public models dedicated to supporting Artemis, with the goal of providing useful information to enhance human presence on the Moon. Campbell Watson, head of research at IBM Research, explains that the opening of the model and dataset aims to provide a common foundation for the global scientific community, adaptable to new questions. He emphasizes that this approach extends a tradition of broad access and collaboration based on decades of public lunar data.

Hugging Face also hosts Prithvi and a 60-year collaboration

This model is not the only open-source contribution resulting from the collaboration between the two entities on Hugging Face, which also offers the Prithvi model series dedicated to meteorology, climate, and Earth observation. IBM and NASA have been collaborating for over 60 years, a partnership that began during the early missions that enabled humans to walk on the Moon.