For the first time, an Earth observation satellite has found what it was looking for — on its own, without human analysts on the ground. The milestone, which occurred in April, marks the first reported use of a vision-language model in orbit, and offers a glimpse of how AI could fundamentally change what space-based sensors are capable of — and how much they’re worth.
Typically, satellites download large chunks of data to analysts on the Earth below, who use machine learning algorithms or their own eyes to figure out what’s going on. But onboard Yam-9, a spacecraft built by space infrastructure company Loft Orbital, a software package built by NASA’s Jet Propulsion Laboratory identified areas of interest in response to natural language queries.
Google DeepMind’s Gemma 3 — the vision-language model, or VLM, that powered the demonstration — is purpose-built for edge applications, meaning it is designed to run on limited hardware far from a data center. VLMs combine the contextual understanding of large language models with the ability to analyze imagery: researchers asked the model to classify sensor data where natural environment meets human development, for example, or to identify infrastructure around railway hubs — and it did.
The demonstration is significant for two reasons. In the near term, it could make space sensors far more useful by doing initial data triage on orbit, reducing the flood of raw data that analysts currently have to wade through. Longer term, it’s a proof point toward running larger-scale AI infrastructure in space.
“It opens the door to always-on, patrol layers in space,” Loft’s head of AI, Paul Lasserre, told TechCrunch. “If you have a VLM, you can have logic—like ‘monitor this border for me, and let me k …