For the first time ever, a satellite found what it was looking for — without a single human analyst on the ground.
The milestone happened in April aboard a spacecraft called Yam-9, built by space infrastructure company Loft Orbital. A software package from NASA's Jet Propulsion Laboratory (JPL) let the satellite identify areas of interest in response to natural language queries.
Normally, satellites just take pictures and beam down huge chunks of data. Analysts on Earth then run machine learning algorithms or use their own eyes to figure out what's going on. That process can take hours or days. Yam-9 did it in orbit, on its own.
The software is a vision-language model — a type of AI that can understand both images and text. It's the same kind of technology behind tools like Google's Gemini or OpenAI's GPT-4 with vision, but it's adapted to run on a satellite's limited computer power.
"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 researchers said.
Why does this matter? Satellites are expensive — both to build and to operate. A big chunk of that cost comes from the data pipeline: downlinking, storing, processing, analyzing. If a satellite can do that work in space, it can send down only the relevant bits. That means faster answers, lower costs, and more value per satellite.
It also opens the door to new kinds of missions. Imagine a satellite that can spot a wildfire while it's still small, or find a missing boat in the ocean, or track illegal fishing in real time — all without waiting for a human to tell it where to look.
Yam-9 is one of their satellites, and JPL's software was the payload that made history.
NASA's Jet Propulsion Laboratory has been pushing AI for space for years. They've used machine learning to analyze Mars rover images, classify galaxies, and predict weather. But putting a vision-language model on an operational satellite is a first.
The test worked — the satellite correctly identified targets based on natural language prompts. The details of what it found haven't been released, but the fact that it worked at all is the breakthrough.
What's next? Loft Orbital and JPL are expected to expand the capability to more satellites. If the tech proves reliable, it could become standard on future Earth observation missions. That would mean more autonomous satellites, less human babysitting, and a whole new way of watching the planet from above.