NASA’s AI Satellite Just Made a Decision Without Humans — in 90 Seconds

NASA is piloting a new form of onboard artificial intelligence that may significantly change how Earth-observing satellites collect and prioritize data. In a recent test, a satellite was able to autonomously detect clouds in its path, process the information onboard, and decide in less than 90 seconds whether to capture or skip a ground image — without any help from mission control.

Dynamic Targeting Helps Satellites “Think”

The technology behind this breakthrough is called Dynamic Targeting, a concept developed over the past decade at NASA’s Jet Propulsion Laboratory in Southern California. It marks a leap toward autonomous spacecraft decision-making.

Steve Chien, the project’s principal investigator and a technical fellow in AI at JPL, explained the ambition behind the effort: “The idea is to make the spacecraft act more like a human. Instead of just seeing data, it’s thinking about what the data shows and how to respond.” The current goal is to allow satellites to distinguish between clear skies and clouds — and skip cloud-obstructed shots that would waste bandwidth and storage.

Cloud-dodging With Onboard Processing

The first flight test was conducted on CogniSAT-6, a CubeSat the size of a briefcase, launched in March 2024. Operated by Open Cosmos and equipped with an AI processor developed by Ubotica, the spacecraft successfully demonstrated the Dynamic Targeting system’s core functionality: detecting and avoiding clouds.

Since the satellite doesn’t have a dedicated forward-looking camera, it tilts 40 to 50 degrees to take images ahead of its orbital path using its optical sensor, which captures both visible and near-infrared light. The onboard AI processes the image using a specialized algorithm trained to identify clouds. If the scene is clear, the spacecraft prepares to image the ground; if it’s cloudy, it cancels the operation to save storage and power.

Ben Smith of JPL, part of NASA’s Earth Science Technology Office that funds the project, highlighted the practical gain: “If you can be smart about what you’re taking pictures of, then you only image the ground and skip the clouds. That way, you’re not storing, processing, and downloading all this imagery researchers really can’t use.”

All of this — from tilting the satellite to analyzing images and adjusting the imaging plan — takes place in just 60 to 90 seconds. Meanwhile, the satellite continues to race around the planet in low Earth orbit at speeds nearing 17,000 mph.

From Avoiding Clouds To Hunting Wildfires

Though the current focus is on cloud avoidance, NASA’s long-term plan is far more ambitious. Upcoming tests will flip the script: instead of avoiding clouds, Dynamic Targeting will seek them out, identifying severe storms and weather systems in real time. Other algorithms will allow the AI to detect thermal anomalies like wildfires and volcanic eruptions, aiming to capture transient phenomena that often elude current satellite systems.

Each of these future use cases will require finely tuned models, capable of identifying specific patterns with enough accuracy to adjust the satellite’s behavior on the fly. Chien called this first successful test “a hugely important step,” setting the stage for future deployments on operational science missions.

Toward Intelligent Satellite Networks

NASA’s vision goes beyond equipping a single satellite with AI. The team is already planning to test a concept called Federated Autonomous MEasurement, which would enable multiple satellites to collaborate. A lead satellite could analyze imagery and communicate targeting instructions to trailing spacecraft, allowing an entire constellation to work together to focus on specific phenomena.

NASA also sees potential for applying Dynamic Targeting in deep space. The team previously experimented with autonomous plume detection using data from ESA’s Rosetta orbiter, targeting emissions from comet 67P/Churyumov-Gerasimenko.

On Earth, this technology could be adapted for radar-based systems to study rare and fast-evolving events like deep convective ice storms, using look-ahead sensing to lock onto these extreme weather patterns as they form. NASA’s broader goal is to deploy agile, responsive instruments that can deliver “novel measurements” across a range of missions.