Satellite Autonomously Detects Target Using Vision Language Model in Orbit

As informed by Techcrunch For the first time, an Earth-observation satellite independently detected the object it was looking for, without the involvement of human analysts on the ground. This achievement, documented in April, marked the first documented use of a Vision-Language Model (VLM) in orbit and gives a sense of how artificial intelligence could fundamentally change the capabilities of space sensors and their value. Usually satellites transmit ground data to analysts, where machine-learning algorithms or a human eye are used to interpret the situation. But on board Yam-9, Loft Orbital's spacecraft, a software package from NASA's Jet Propulsion Laboratory identified areas of interest in response to natural-language queries. Gemma 3 from Google DeepMind is the very Vision-Language Model that was used in the demonstration. It is designed for edge devices, meaning it can run on limited hardware outside data centers. The VLM combines contextual understanding of large language models with the ability to analyze images: researchers asked the model to classify sensor data where nature meets human development, or to detect infrastructure around railway hubs – and it handled it. The demonstration matters for two reasons. First, in the near term it could make space sensors much more useful by performing on-orbit data preprocessing and reducing the stream of raw data to be analyzed on Earth. Second, it is a proof of concept for the possibility of launching a more scalable AI infrastructure in space. 'If you have a VLM, you can have logic – for example, monitor this boundary for me and report when something is suspicious,' and interact with satellites. – Paul Lasserre Technical Details and Next Steps Loft develops its craft as platforms for external clients. The business model is closer to infrastructure as a service than to traditional satellite manufacturing. A recent deal envisions creating, launching, and operating six new satellites for EarthDaily that will analyze and sell the data collected on board. Yam-9 was launched in fall 2025 as a flagship for Loft's AI policy and includes an Nvidia Jetson Orin AGX graphics processor – one of the leading chips used in space computing. Juan Delfa Victoria, the technical lead of the AI group at NASA JPL, led the development of NAVI-Orbital – the software package that effectively became the 'chassis' for Gemma 3 VLM. Although Gemma 3 is a ready-made solution, the developers simplified the package, reducing the number of libraries and memory requirements. Although this is the first documented use of a VLM on orbit, others are expected to follow suit. Planet Labs uses satellites with Jetson Orin processors; they are currently used for simpler object-detection tasks, but research is underway into other AI applications, including VLM. Kepler Communications, which operates the largest GPU fleet in space, declined to say whether it has deployed a VLM due to NDA with partners, but noted that since the launch of its spacecraft there have been several undisclosed instances of using our computing environment. 'Now that the concept is proven, it's truly a milestone in motion,' said Lasserre. 'The goal is to build a constellation with real-time coverage of any point on Earth, which, according to an official, would require from 50 to 100 Yam-9-like satellites. Loft currently operates 12 spacecraft in orbit.' Lessons learned from deploying these smaller models on orbit will help explain how companies might roll out larger-scale computing infrastructure in space in the future, especially for everyday but critically important issues of power and memory management. It could also open the door to new scientific tools. The NAVI-Space idea originated in the mind of JPL researcher Taran Cyriac John, who was thinking about digital assistants for astronauts exploring the Moon or Mars. 'We think astronauts need assistance, as in video games and movies, where artificial intelligence becomes an interactive helper,' noted Delfa Victoria. 'And it's not to be called HAL 9000,' they add, in the context of caution about naming a general-purpose system. Future perspectives suggest that VLM technologies could fundamentally change not only how data from space platforms is processed but also prepare new approaches to science and research in orbit and beyond. Coordinated development pace points to the need for more attention to practical and energy-efficient computing solutions in space, where resources are as valuable as the data from the stars. Looking to the future, the first wave of VLM use on orbit could spur the creation of more autonomous space systems that can understand the context of imagery on their own and make decisions without constant ground control.