Satellite Now Finds Objects Autonomously, Marking Advance in Space AI

For the first time, an Earth-observing satellite found its target independently, without human analysts on the ground. This April milestone marked the first reported use of a vision-language model in orbit, offering a glimpse of how AI could fundamentally reshape the capabilities and value of space-based sensors.

Typically, satellites send large volumes of data to analysts on Earth, who rely on machine learning or manual inspection to interpret it. But onboard Yam-9, a spacecraft built by Loft Orbital, a software package from NASA’s Jet Propulsion Laboratory identified areas of interest in response to natural language queries.

Google DeepMind’s Gemma 3, the vision-language model (VLM) powering the demonstration, is designed for edge applications—running on limited hardware far from data centers. VLMs combine the contextual reasoning of large language models with image analysis capabilities. For example, researchers asked the model to classify sensor data where natural environments meet human development, or to identify infrastructure near railway hubs, and it delivered.

This demonstration matters for two reasons. In the near term, it could make space sensors far more useful by performing initial data triage in orbit, reducing the flood of raw data analysts currently have to wade through. Longer term, it serves as a proof of concept for running larger-scale AI infrastructure in space.

“It opens the door to always-on patrol layers in space,” Loft’s head of AI, Paul Lasserre, told TechCrunch. “With a VLM, you can implement logic—such as ‘monitor this border and alert me when something looks suspicious’—and interact back and forth with the satellites.”

Loft’s spacecraft are built as platforms for third-party customers. The business model resembles infrastructure-as-a-service more than traditional satellite manufacturing. In a recent deal, Loft built, launched, and operated six new satellites for EarthDaily, which will analyze and market the data collected onboard. Yam-9 launched in fall 2025 as a pathfinder for the company’s orbital AI projects, and it carries an Nvidia Jetson Orin AGX GPU, one of the leading chips used in space computing.

Juan Delfa Victoria, a technical leader in NASA JPL’s AI group, led the development of NAVI-Orbital, a software package that served as the harness for the Gemma 3 VLM. Although Gemma 3 is off the shelf, software engineers had to streamline the package to reduce the libraries and memory required.

While this is the first reported use of a VLM in orbit, other companies are expected to follow. Planet Labs operates satellites with Jetson Orin processors, currently using them for simpler object detection tasks. However, a spokesperson says research is underway on other AI applications, including VLMs.

Kepler Communications, which operates the largest cluster of GPUs in space, declined to comment on whether it has deployed VLMs in orbit, citing NDA agreements with partners. However, it noted that there have been “several undisclosed use cases of our compute environment” since those spacecraft launched in January.

“Now that we’ve proven the concept, that’s the direction we’re heading,” Lasserre said. The goal is to expand the constellation to provide real-time coverage anywhere on Earth, which he says would require 50 to 100 satellites like Yam-9. (Loft currently operates 12 spacecraft in orbit.)

Lessons from deploying these smaller models in orbit will inform how companies approach larger-scale compute infrastructure in space, especially in the prosaic but critical areas of power and memory management.

They could also pave the way for new scientific tools. The concept for NAVI-Space originated with JPL researcher Taran Cyriac John, who envisioned digital assistants for astronauts exploring the Moon or Mars.

“We’re thinking: you have astronauts in pressurized suits, and they can’t tap 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 movies, where you see an AI that is interactive?”

Just don’t call it HAL 9000.