source: techcrunch ai: a satellite just learned to find things on its own — here’s what that means
level: technical
in april, yam-9, a satellite built by loft orbital, used a vision-language model from google deepmind called gemma 3 to identify areas of interest on its own. the software, developed by nasa's jet propulsion laboratory, let operators ask the satellite to find things like infrastructure near railway hubs or where nature meets human development. this is the first reported use of a vlm in orbit, moving beyond simple object detection to understanding complex scenes from natural language prompts.
the demo shows how ai can cut the flood of data satellites send to ground analysts. by doing initial triage in space, the model can flag only relevant images, saving bandwidth and analyst time. loft orbital sees this as a step toward always-on monitoring, where satellites can watch borders or other areas and alert humans only when something suspicious appears. the company plans to build a constellation of 50 to 100 such satellites for real-time global coverage.
the yam-9 satellite carries a nvidia jetson orin agx gpu, a chip designed for edge computing. engineers had to slim down the software to fit the satellite's limited memory and power. other companies like planet labs and kepler communications are also exploring vlms in orbit, though details are scarce. the technology could eventually support digital assistants for astronauts on deep-space missions, helping them interact with their environment through voice commands.
why it matters: onboard ai reduces data bottlenecks and enables faster, autonomous decision-making for earth observation, making satellite data more actionable and cost-effective.
source: techcrunch ai: a satellite just learned to find things on its own — here’s what that means