NASA Sends Google Gemma 3 Language Model to Orbit

NASA's Jet Propulsion Laboratory sent Google's Gemma 3 to space, achieving the first in-orbit demonstration of a vision-language model analyzing imagery from a satellite's own sensor. The system, known as NAVI-Orbital, used Gemma 3 to analyze images captured by a YAM-9 satellite built by Loft Orbital. NAVI-Orbital is an agentic software framework developed by Juan M. Delfa and coauthors that deploys a compressed 4-bit format of Gemma 3 4B to produce plain-text image descriptions. In ground-based benchmarks, it achieved 88% accuracy on 7,960 images without fine-tuning. In orbit, NASA performed two live capture tests over Toulouse, France, and the coast of Argentina, where Gemma 3 generated text descriptions and answered scripted questions. The model ran on Nvidia's Jetson Orin AGX, requiring only 8 GB of memory. The project demonstrates semantic compression, allowing satellites to downlink text summaries instead of large image files, reducing data from megabytes to kilobytes. This could enable faster detection of events like wildfires, reducing delays from up to 90 minutes to near real-time. NAVI-Orbital is walled off from flight software but provides a proof of concept for interacting with spacecraft via natural language prompts, potentially assisting astronauts in future missions.