Gemma 3, a vision-language model designed for deployment on constrained hardware, combines image analysis with language-based instructions. During the demonstration, researchers directed the model to identify areas where developed infrastructure intersects with natural landscapes and to locate infrastructure associated with railway hubs. According to the teams involved, the system successfully completed those tasks.
The software environment supporting the model, known as NAVI-Orbital, was developed by NASA JPL’s AI group under the leadership of Juan Delfa Victoria. Although the underlying model was commercially available, engineers modified the software stack to reduce memory requirements and limit the number of supporting libraries needed for operation in orbit.
Loft sees broader implications beyond data filtering. Paul Lasserre, the company’s head of AI, said the technology could enable satellites to continuously monitor specific locations and respond to user-defined objectives. “It opens the door to always-on, patrol layers in space,” Lasserre told TechCrunch. “If you have a VLM, you can have logic—like ‘monitor this border for me, and let me know when something is suspicious,’ and interact back and forth with the satellites.”
The demonstration also serves as a step toward more advanced computing infrastructure beyond Earth. Yam-9 was launched as a test platform for Loft’s orbital AI efforts and carries an Nvidia Jetson Orrin AGX GPU, a processor commonly used in space-computing applications. Loft’s business centers on providing satellite infrastructure for external customers, including a recent agreement to build, launch, and operate six spacecraft for EarthDaily.
Interest in onboard AI is spreading across the sector. Planet Labs already operates satellites equipped with Jetson Orin processors and says it is researching additional AI capabilities, including vision-language models. Kepler Communications, which operates a network of space-based GPUs, declined to discuss specific deployments because of partner confidentiality agreements but said multiple undisclosed applications have run on its orbital computing platform.
For Loft, the successful demonstration marks the beginning of a larger effort. Lasserre said the company ultimately aims to deploy enough satellites to provide real-time coverage across the globe, estimating that such a network would require between 50 and 100 spacecraft similar to Yam-9. Loft currently operates 12 satellites.
Beyond commercial applications, the technology could support future scientific and exploration missions. The concept behind NAVI-Space, the broader initiative that led to NAVI-Orbital, originated from work by JPL researcher Taran Cyriac John, who explored how AI assistants might support astronauts during lunar or Martian missions.
“We’re thinking, okay, you have astronauts with pressurized suits, and you know they cannot be tapping on a keyboard, whatever they want to do is complex,” Delfa Victoria said. “So, how about we provide an assistant, like in video games and in movies, where you see an AI which is interactive?”
The experiment suggests that AI systems capable of understanding both images and language may increasingly become part of space operations, handling tasks directly in orbit while reducing reliance on constant ground-based analysis.
This analysis is based on reporting from TechCrunch.
Image courtesy of NASA/USGS.
This article was generated with AI assistance and reviewed for accuracy and quality.
