WASHINGTON — Loft Orbital is working with NASA’s Jet Propulsion Laboratory to test the use of artificial intelligence on spacecraft to improve Earth science monitoring.
Loft Orbital announced June 23 it has an agreement with JPL to test the lab’s AI software on Loft spacecraft as part of a NASA-funded project called Federated Autonomous Measurement, or FAME.
The tests started this month using a Loft Orbital spacecraft whose onboard computers are running the JPL software. Additional tests are planned in 2027 and 2028 using future Loft Orbital spacecraft.
The goal of the effort is to automate a process known as tip-and-cue, where imagery from one spacecraft is used to identify features to be observed in greater detail or using other techniques by another spacecraft. That process today typically requires imagery to be downlinked and analyzed on the ground.
“You can use different assets with processing at the edge to capture, sense, understand and send insights about what’s going on on Earth without having to downlink big amounts of data,” Paul Lasserre, general manager for AI at Loft Orbital, said in an interview.
That approach uses AI models trained on a “very large corpus” of data, he said. “It can recognize everything without being told what to look for,” he said.
The project has two technical challenges. One is to combine the infrastructure on the satellite, including both sensors and processors that can take in and process images in real time. The other is the availability of open-source AI models small enough to operate within the hardware constraints of the satellite. Only now has the combination of space hardware and AI models made this feasible.
“It’s the intersection of the AI state of the art that started to make smaller models that were high performance with multimodal reasoning, which six months ago were not a thing, and the infrastructure that has everything it takes,” he said, “and someone like NASA that had this vision for some time and identified that now was the time these different elements were coming together.”
Lasserre, whose background is in AI in terrestrial applications, said he had to reconsider the capabilities of AI in space given the computing limitations on spacecraft. “The constraint exists, but it’s not really limiting,” he said. “We can run state-of-the-art models if we have prepared them well.”
He said one vision of how this would work involves one spacecraft with a sensor that is always on, collecting data and using the AI model to look for features of interest, flagging them and transmitting details to other spacecraft through intersatellite links for rapid follow-up. “You can have this ‘patrol mode’ that was not really a thing before” because of processing and connectivity bottlenecks, he said.
Examples he provided for this approach include being able to quickly detect and follow up on wildfires or marine pollution events. “You can imagine all of the security, military and intelligence applications where you want to be able to have the information rapidly to trigger some actions,” he added.
Loft Orbital plans to expand its capabilities with a series of 10 satellites called Altair with multiple sensors, edge computing for AI models and intersatellite links.
“There is no point in having all of this value chain of real-time insights and autonomy if you need to wait hours for the next satellite,” he said. “To me, that’s the tipping point, where the value is just suddenly a lot higher commercially and for governments.”
