DARPA project uses AI to warn of space weapons and spy satellites

Applications of AI


As more governments and private companies turn to proliferating constellations of satellites for added capacity, some defense experts worry that these large fleets could be used to hide space weapons or spy satellites.

The Defense Advanced Research Projects Agency effort aims to use artificial intelligence to uncover potentially nefarious capabilities.

In 2023, DARPA selected California-based space technology company Slingshot Aerospace to develop an AI system to identify anomalous satellites within these massive constellations. The company unveiled a model, named Agatha, on June 5 and said it had demonstrated the ability to detect anomalous satellites within operational constellations.

“Identifying malfunctioning or potentially malicious objects and their intent within a large satellite constellation is a complex challenge that required us to go beyond traditional approaches and develop novel, scalable AI algorithms,” Dylan Kessler, director of data science and AI at Slingshot, said in a statement. “Our Agatha model has also proven capable of delivering high-quality insights that provide 'explainability,' or context for why a particular object was flagged.”

China has announced plans to launch two giant constellations of tens of thousands of satellites over the next few years, as a challenge to Elon Musk's SpaceX, whose Starlink communications satellite network includes more than 6,000 operational spacecraft.

At the same time, Pentagon officials have confirmed that Russia is developing satellites capable of carrying nuclear weapons, and last month launched a counter-space weapon designed to track U.S. spy satellites.

Audrey Schaffer, vice president of strategy and policy at Slingshot, told C4ISRNET that these developments make tools like Agatha especially important.

“Agatha is an example of technology designed to help clarify vision in increasingly crowded and congested environments,” she said. “In particular, Agatha can really help you find the needle in the haystack.”

To train Agatha's algorithms, Slingshot ingested more than 60 years of data from simulated megaconstellations that the company created. These constellations also contained anomalous satellites, allowing Agatha to distinguish between types of satellites and flag anomalies.

The system incorporates a technique called inverse reinforcement learning, which allows Agatha not only to recognize and track anomalous spacecraft maneuvers and other activities, but also to assign motivations to those actions.

“AGATHA can do more than just identify that this particular satellite is anomalous,” Schaffer said, “it can also assess why this satellite is behaving differently than the others in the constellation, and what policies or operational directives explain that difference in behavior.”

Once the model was trained, Slingshot began testing Agatha's capabilities using existing commercial satellite constellations. Schaffer wouldn't say which companies' satellites the system observed, but he said it flagged some benign anomalies and then checked in with the spacecraft owners.

The DARPA-led project wrapped up earlier this year, and Schaffer said the company is currently in talks with government and commercial customers that may be interested in Agatha. He noted that the tool could be useful to the U.S. Space Command and the National Space Defense Center, which monitor activity in space.

“Space traffic is only going to grow,” she says. “When you have not just 10,000 operational satellites, but 10,000 satellites in a constellation, it's quickly going to be impossible for a human, or even a team of humans, to sift through all that data and identify potential threats to national security.”

Courtney Albon is a space and emerging technology reporter for C4ISRNET. She has covered the U.S. military since 2012, focusing on the Air Force and Space Force. She has reported on some of the Department of Defense's most significant acquisition, budget and policy challenges.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *