Prithvi AI model deploys in orbit on Kanyini satellite and ISS payload
Updated
Updated · mitsloanme.com · May 8
Prithvi AI model deploys in orbit on Kanyini satellite and ISS payload
10 articles · Updated · mitsloanme.com · May 8
Adelaide University and SmartSat CRC compressed and uploaded the NASA-IBM open-source model, trained on 13 years of Landsat and Sentinel-2 data, to South Australian and ISS platforms.
The in-orbit test aims to bypass bandwidth limits by letting one foundation model handle tasks including flood mapping, cloud detection, disaster monitoring and crop-yield prediction.
Researchers say the milestone could lead to conversational satellite operations and wider use of foundation models in planetary science, astrophysics, and biological and physical sciences.
How was a massive AI model shrunk for orbit without losing its power to predict floods or monitor global crops?
Now that satellites can think, how soon will they autonomously respond to disasters on Earth without human command?
As open-source AI watches Earth from space, who governs its use and prevents its misuse for global surveillance?
In August 2024, NASA and IBM, together with Australian partners, achieved the first in-orbit deployment of Prithvi, a versatile geospatial AI model capable of real-time Earth observation on small satellites like the Kanyini CubeSat. This breakthrough was enabled by advanced model compression and a unique update system allowing efficient in-space adaptation. Prithvi processes complex hyperspectral data onboard, drastically reducing data transmission and accelerating disaster detection, such as wildfire smoke alerts. Its open-source design fosters global collaboration and continuous improvement through user feedback. Looking ahead, NASA and IBM plan to expand AI models for planetary science and enable natural language interaction with spacecraft, marking a new era of intelligent, autonomous space exploration.