en.Wedoany.com Reported - NASA's Jet Propulsion Laboratory, in collaboration with space infrastructure company Loft Orbital, successfully tested a visual language model (VLM) on the YAM-9 satellite. The model can respond to natural language queries and automatically identify areas of interest on the ground. The demonstration used Google DeepMind's Gemma 3 model, which is designed for edge applications and can run on limited hardware far from data centers, combining the contextual understanding capabilities of large language models with image analysis.
During the test, researchers asked the model to classify sensor data at the intersection of natural environments and human development, or to identify infrastructure around railway hubs, and the model completed the tasks. The YAM-9 satellite, built by Loft Orbital and launched in fall 2025, serves as a pathfinder for the company's orbital AI project. It is equipped with an Nvidia Jetson Orin AGX GPU, one of the leading chips used in space computing. Loft's spacecraft are designed as platforms for third-party customers, with a business model closer to infrastructure as a service.
The significance of this demonstration lies in the short term: by performing initial data classification in orbit, it reduces the vast amount of raw data that analysts need to process. In the long term, it validates the feasibility of operating larger-scale artificial intelligence infrastructure in space. Paul Lasserre, Loft's AI lead, stated that having a VLM enables continuous monitoring and logical interaction, such as monitoring borders and notifying users when suspicious situations arise.
Juan Delfa Victoria, technical lead of the NASA JPL AI group, led the development of the NAVI-Orbital software package, which serves as the "framework" for the Gemma 3 VLM. Engineers had to streamline the software package to reduce the required libraries and memory, enabling it to operate in orbit. This is the first reported use of a VLM in orbit, but other companies may follow. Planet Labs' satellites are equipped with Jetson Orin processors, currently used for simple object detection, but a spokesperson said the company is researching other AI applications, including VLMs. Kepler Communications declined to disclose whether it has deployed VLMs due to confidentiality agreements, but noted several unannounced computing environment use cases since its launch in January.
Lasserre said that after validating the concept, the future direction is to expand the constellation to ensure real-time coverage of any location on Earth, which would require approximately 50 to 100 satellites like the YAM-9 (Loft currently operates 12 spacecraft in orbit). Experience with deploying smaller models in orbit will influence the deployment of larger-scale computing infrastructure, particularly in the areas of power and memory management. The NAVI-Space concept originated with Delfa Victoria and JPL researcher Taran Cyriac John, who initially envisioned it as a digital assistant for astronauts exploring the Moon or Mars, allowing astronauts to interact with AI via voice while wearing pressurized suits.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
