NASA testing AI to make satellites smarter

NASA’s Jet Propulsion Laboratory (JPL) in Southern California is testing a new artificial intelligence-driven technology known as Dynamic Targeting, which enables satellites to autonomously decide where to focus their scientific observations in real time. This innovation allows spacecraft to analyze their surroundings and respond within seconds, significantly improving the quality and efficiency of Earth data collection.

Image credit: NASA/JPL-Caltech

NASA testing AI to make satellites smarter: Smarter observations, less cloud interference

In mid-July 2025, NASA conducted the first in a series of flight tests demonstrating how Dynamic Targeting enables Earth-observing satellites to anticipate their orbital path, analyze imagery onboard using AI, and autonomously determine optimal points of interest for observation. This process takes under 90 seconds and operates without human intervention.

The primary goal of the test was to demonstrate the technology’s ability to avoid cloud cover, a persistent challenge in satellite-based Earth imaging. Clouds obstruct optical sensors on satellites up to two-thirds of the time, often rendering collected imagery unusable. Dynamic Targeting addresses this by looking ahead approximately 300 miles (500 kilometers) to assess upcoming atmospheric conditions. If clear skies are detected, the satellite proceeds to image the ground. If clouds are identified, it skips the imaging to conserve storage and prioritize usable data.

Ben Smith, an associate at NASA’s Earth Science Technology Office, emphasized the importance of this feature, stating that Dynamic Targeting helps researchers avoid collecting and processing unusable images, thereby increasing the efficiency of Earth observation missions.

How the technology works

The tests are being conducted aboard CogniSAT-6, a compact CubeSat launched in March 2024. The satellite is operated by Open Cosmos and equipped with a payload developed by Ubotica, which includes a commercial AI processor. Prior to this, similar AI algorithms were tested aboard the International Space Station using the same type of processor, affirming their viability for orbital operations.

Since CogniSAT-6 does not include a dedicated look-ahead imager, it physically tilts forward 40 to 50 degrees to use its primary optical sensor to capture forward-looking imagery in visible and near-infrared light. Once this data is collected, Dynamic Targeting's onboard algorithm, trained to detect clouds, analyzes the image. The planning software then selects clear-sky targets, prompting the satellite to return to a downward-facing position to capture surface images. This entire sequence is completed within 60 to 90 seconds, even as the satellite travels at nearly 17,000 mph (7.5 km/s) in low Earth orbit.

Future applications and testing

Following the successful demonstration of cloud avoidance, future tests will shift focus to seeking out and imaging transient events such as wildfires, volcanic eruptions, and extreme weather phenomena. The JPL team has developed specialized algorithms tailored to identify and track these brief but critical events.

Principal investigator Steve Chien, a JPL technical fellow in AI, described the ultimate aim of the project as creating highly responsive, intelligent instruments capable of novel, adaptive measurements. The long-term vision includes deploying Dynamic Targeting on science missions both in Earth orbit and beyond. Chien noted that inspiration for this technology partly came from a prior project that used data from ESA’s Rosetta mission to autonomously detect and image plumes from comet 67P/Churyumov-Gerasimenko.

On Earth, Dynamic Targeting could be combined with radar systems to observe rare weather events like deep convective ice storms, which are too fleeting to be reliably captured with current technologies. Algorithms would identify these ice storms using forward-looking instruments, prompting radar-equipped satellites to pivot and focus their instruments, collecting detailed data over extended viewing times.

Additionally, NASA is exploring coordinated AI use across multiple satellites. In this federated model, one satellite could detect a target and then alert trailing satellites to focus on the same phenomenon. This concept, known as Federated Autonomous MEasurement, will undergo initial testing later in 2025 under Chien’s leadership.

How JPL’s Dynamic Targeting uses a lookahead sensor to see what’s on a satellite’s upcoming path

Image credit: NASA/JPL-Caltech