Loft Orbital, in partnership with NASA’s Jet Propulsion Laboratory (JPL), has initiated testing of a new artificial intelligence system for satellites. This technology will enable spacecraft to independently detect significant events on Earth and instantly relay that information to other satellites, eliminating the need to transmit large volumes of data back to ground stations.
Loft Orbital announced the start of these trials on June 23. The project is being carried out under NASA’s Federated Autonomous Measurement (FAME) program, which aims to develop distributed autonomous systems for Earth observation. The first phase of testing kicked off in June aboard one of the company’s operational satellites, where JPL’s software is running directly on the satellite’s onboard computing systems. Additional tests are scheduled for 2027 and 2028, utilizing new satellites from Loft Orbital.
The core objective of the project is to automate a process known as tip-and-cue. Currently, a satellite typically captures an image of Earth’s surface, transmits that data to the ground for analysis by specialists, and then, if needed, sends commands to other spacecraft to conduct more detailed observations of the same object or event. This cycle can waste considerable time.
The system under development aims to shift most of this work directly into orbit. Artificial intelligence will analyze images immediately after capture, identify objects or phenomena of interest, and automatically alert other satellites to immediately perform follow-up observations using different sensors or imaging methods.
According to Loft Orbital’s head of artificial intelligence, Paul Lasserre, this new architecture will make it possible to obtain useful insights without transmitting massive amounts of raw data to Earth.
The system employs AI models trained on extremely large datasets. Rather than searching for pre-defined objects, these models can independently recognize a broad range of phenomena and anomalies on the planet’s surface.
To bring the project to life, two complex technical challenges had to be addressed simultaneously. The first relates to the satellite’s hardware: each spacecraft must carry both modern sensors and sufficiently powerful computing systems capable of processing images in real time. The second challenge involves the AI models themselves. Until recently, they were too large and demanded too much computing power to function in space.
According to the developers, the situation has changed only in the past few months, thanks to the emergence of a new generation of compact models capable of performing so-called multimodal reasoning—analyzing and cross-referencing information from various data sources while using significantly fewer computing resources. It is the combination of such models with modern satellite computing platforms that has made the project practically achievable.
Lasserre noted that his experience with AI systems on Earth initially made him skeptical about the potential of similar technologies in space, due to the limited performance of satellite computers. However, modern approaches have shown that these constraints are no longer critical.
One envisioned scenario for the system works like this: A single satellite continuously monitors a specific region of Earth, acting as a sort of orbital patrol. Its AI constantly analyzes incoming images, and upon detecting a suspicious object or event, it automatically relays information to other spacecraft via inter-satellite communication links. These satellites can then immediately perform more detailed observations of the target, without waiting for input from ground operators.
Developers cite examples such as rapid detection of wildfires, monitoring of ocean surface pollution, tracking of environmental incidents, and other fast-evolving events. Beyond civilian tasks, the technology also has potential applications in security, defense, and intelligence, where the speed of information delivery is often decisive.
To further advance this concept, Loft Orbital plans to build the Altair constellation, consisting of ten next-generation satellites. Each spacecraft will be equipped with multiple types of sensors, computing systems to run AI models directly onboard, and inter-satellite communication links for fast data exchange.
According to the company, it is the combination of autonomous data analysis, inter-satellite coordination, and persistent monitoring that will enable a shift from the traditional image-collection model to a system that essentially decides for itself which events require immediate attention. This could significantly increase the value of satellite data for both government agencies and commercial clients.
If the trials prove successful, future satellite constellations will not merely photograph Earth and send data for human analysis—they will independently detect fires, pollution, natural disasters, and other critical events in near-real time. For space-based observation systems, this would mark one of the most significant steps toward fully autonomous operations in orbit.
About the Author
