YAM-9 Satellite Identifies Image Features Onboard With 88.2% Accuracy
Updated
Updated · ScienceAlert · Jul 1
YAM-9 Satellite Identifies Image Features Onboard With 88.2% Accuracy
1 articles · Updated · ScienceAlert · Jul 1
Summary
YAM-9 has identified and described features in its own image scans without ground control, letting operators use natural-language prompts such as requests to find railway hubs or flood damage.
NAVI-Orbital—built by NASA JPL and Loft Orbital—runs Google DeepMind's Gemma 3 onboard, replacing custom command sequences with editable prompts and cutting bandwidth, power and retasking delays.
7,960 baseline test images were classified at 88.2% accuracy across categories including residential areas, beaches, farmland and mountains, though only two live in-orbit captures have been completed so far.
100 satellites of this type could enable near-real-time global monitoring for wildfires, ports, borders and oil spills, while raising surveillance, reliability and adversarial-prompt concerns the researchers say remain unresolved.
As AI satellites accept simple text commands, how can we prevent them from being hijacked by malicious prompts?
Could this on-demand intelligence from private companies make government-owned spy satellites obsolete?
When a private AI can monitor the globe in real-time, who controls what it sees and how it interprets reality?
From Data Collector to Autonomous Analyst: YAM-9’s Historic 2026 AI Demonstration Redefines Satellite Operations
Overview
In April 2026, Loft Orbital made history by running Google DeepMind’s Gemma 3 vision-language model directly on its YAM-9 satellite. This breakthrough transformed YAM-9 from a simple data collector into an autonomous analyst, able to answer natural language queries, classify Earth imagery, and generate plain-English summaries—all in orbit, without sending raw data to Earth first. This shift fundamentally changed how Earth observation satellites operate, showing that satellites can process and analyze data themselves, making space missions faster, more efficient, and paving the way for smarter, more responsive space technology.