AI-Powered YAM-9 Satellite Can Locate Objects on Earth via Text Commands

Space-based remote sensing technology has made history after the YAM-9 satellite successfully identified and described various objects on the Earth’s surface independently, without assistance from ground-based analysis. This achievement marks the first time an artificial intelligence (AI) model capable of understanding language and reading images simultaneously has been operated directly in space. Operating a satellite, which previously required writing complex command sequences, is now simplified to merely entering simple text commands in everyday language. The YAM-9 satellite was developed by space company Loft Orbital and launched in autumn 2025 as a test vehicle for an in-space AI project. The satellite serves as a platform for the NAVI-Orbital system, software resulting from a collaboration between NASA’s Jet Propulsion Laboratory (JPL) and Loft Orbital. Embedded within the satellite are an Nvidia Jetson Orin AGX graphics card and Google DeepMind’s Gemma 3 vision-language model. This model is designed to be compact enough to process text and images simultaneously on hardware with limited capacity, far from data centres. The NAVI-Orbital system is designed with a multi-agent architecture consisting of three modules: a task coordinator, an image analyser, and a dialogue agent that answers operator questions. In the first orbital trial conducted last April, the model successfully classified sensory data, ranging from mapping residential areas, farmland, coastlines, and mountains, to infrastructure around railway hubs. Prior ground testing on 7,960 images also recorded a classification accuracy rate of 88.2 percent. The application of AI directly in orbit is predicted to revolutionise the space industry market by curbing the surge of raw data that typically must be sent to Earth for analysis. To monitor the entire Earth’s surface in real-time, a constellation of around 50 to 100 satellites similar to YAM-9 would be required. “This AI truly ‘sees’ what is in the image and accurately identifies what the analyst is looking for—bridges, highways, specific bodies of water, or signs of natural disasters,” said Sarah Preston, senior marketing manager at Loft Orbital. The development of this technology also opens up opportunities for real-time Earth monitoring. Loft Orbital’s Head of AI, Paul Lasser, likened this capability to a constant patrol from space where the satellite only transmits data when it finds the criteria being sought, such as oil spills, construction on borders, or signs of floods and forest fires. Beyond Earth observation, the NAVI-Orbital technology is designed for future exploration missions to the Moon and Mars. The initial idea for the project emerged when JPL researchers Juan Delfa Victoria and Taran Cyriac John were contemplating a digital assistant for astronauts. In conditions where wearing a high-pressure space suit makes typing on a keyboard impossible, astronauts could give voice commands directly to an interactive AI assistant accompanying their mission. Although the system’s robustness against malicious prompt attacks still requires further research, the development team is optimistic that this orbital AI system will soon become a new standard for space platforms.