Think tanks typically just write reports, but when it came to studying the use of artificial intelligence to help the Space Force track objects in orbit and warn about possible collisions, the RAND Corp. went one step beyond that.
A team of information scientists at RAND actually built AI tools which Space Force analysts could use to “fight tonight,” team leader Li Ang Zhang told Air & Space Forces Magazine.
“Having this kind of AI solution can really benefit the mission today,” he said, alleviating technology bottlenecks and providing a bridge capability until a delayed modernization program for space domain awareness (SDA) kicks in. Space Force officials seem to agree.
Machine learning AI (AI/ML) is quite different from the generative AI large language models that have captured headlines and public imagination in the last two years, but it is vital to help human analysts sift through and make sense of the huge amount of data coming off of and about the rapidly proliferating number of satellites, especially in low-earth orbit, or LEO, said Rudolph “Reb” Butler, a senior advisor to the Space Force’s Chief Technology and Innovation Officer.
The Space Force is working with its partners to ease the cognitive burden on humans dealing with SDA data through automation, he said at a recent CyberSat event: “There’s a lot of work to be done.”
To help with that work, the RAND team went deep on a single use-case for AI/ML called conjunction assessment: the process of identifying and tracking objects in orbit to predict possible collisions.
Conjunction assessment, or CA, is a good way to demonstrate the value of AI/ML right now, because it’s become so much harder in the years since the Space Force was stood up in 2019, Zhang said. More satellites had been launched in the last five years than in the previous 60, he noted.
“We’re having almost weekly launches around the globe, but the number of people doing this mission and the number of computers are relatively static,” he said.
Space Force analysts in Space Defense Squadrons 18 and 19, charged with the CA mission, were in danger of becoming overwhelmed by the growing scale and complexity of their task, Zhang warned.
The Space Force tracks nearly 45,000 objects in orbit, he said. “If you think about the number of possible pairwise comparisons it quickly gets quite staggering,” he said. Add in complexities like atmospheric conditions and solar weather, variations in the Earth’s gravitational field, and the calculations required to predict orbits and warn about possible collisions quickly outpace the available computational power.
That’s especially true, he added, because of the aging legacy technology upon which the Space Force relies. The computers that carry out the orbital calculations are “very old and at capacity, and have been for at least 10 years now,” he said, calling it “a well-known problem, and … one of the main bottlenecks that we have today” in SDA capabilities.
Almost three years ago, Space Systems Command, the part of Space Force responsible for most of its acquisition programs, announced a $49.7 million modernization effort dubbed the Advanced Tracking and Launch Analysis System (ATLAS.) But in a report earlier this year, the Pentagon’s top weapons tester said the effort is two years behind schedule.
Until the modernization effort is complete, Zhang said, the AI tools his team developed could provide a bridging function—allowing the legacy computers to keep handling the growing volume of data and calculations needed to keep up with the expanding and increasingly dynamic orbital environment.
“We want AI to try to bridge that gap in resources for now,” he said. While the legacy computers could keep doing the demanding orbital calculations “in the background,” AI tools which provide a “much, much faster, but a little less accurate” predictions could be used for a “first cut,” to triage analysts’ work, and let them focus their attention and computing power on the most urgent tasks, Zhang said.
Although the legacy computers operate within the classified domain and the RAND team was not able to test their AI/ML tools on them, Zhang said they had been designed and tested on “relatively old machines,” and he was confident they would work on the legacy technology being used for SDA.
To the Space Force’s credit, he said, “they have the infrastructure to bring new algorithms, new cool stuff, especially from small businesses, into their processes. So a lot of the architecture [to import and use these AI/ML tools] does exist. I think it just is a matter of pulling the trigger.”
He said his team was engaging with Space Force to figure out next steps, but declined to comment further. Space Systems Command didn’t respond to requests for an interview or comment.
The situation is complicated by the bureaucratic context, said retired Space Force Col. Charles Galbreath. The part of the SDA mission concerned with CA—the orbital tracking and cataloging of space objects, which he called the “traffic management role”—is being shifted to the Department of Commerce.
The Space Force wants to pass that mission over because it is seen as more suited to a civilian agency given the growing prevalence of commercial activity in orbit. Galbreath also noted that the Space Force wants to focus on newer capabilities it will need as the space domain becomes not just more crowded, but more dynamic, with more satellites able to maneuver themselves in orbit.
To meet those new challenges and prepare to fight a shooting war in orbit if needed, USSF is leaning forward on incorporating new technologies like AI and cannot afford to be gun shy about them, Galbreath said. His last active duty assignment was as the Space Force’s deputy chief technology and innovation officer, and he is now a senior fellow for space studies at the Mitchell Institute for Aerospace Studies.
“The number of objects in space has just been going off the charts,” he said, and would continue to grow as new megaconstellations like Starlink and future competitors such as Amazon’s Kuiper continue to be fielded. “So the risk of not using AI/ML is an outdated system increasingly incapable of producing conjunction warnings that are accurate in a timely fashion. And therefore the risk of an unforeseen or undetected collision increases,” he said.
Retired Maj. Gen. Kim Crider, the Space Force’s former Chief Technology and Innovation Officer, said that while the SDA mission is one of the most urgent use-cases for AI/ML in the service, it is far from the only one.
“It’s an urgent challenge because understanding the domain is foundational to everything else that we’re going to do there,” said Crider, who left the service in 2021 and last year helped found Elara Nova, a strategic consultancy focused on space. Without accurate and constantly updated information about objects in space, “we’re going to have a very hard time making decisions about what to do in the domain and where to apply our defensive effects, or where to engage if we need to engage,” she said.
But the Space Force shouldn’t stop with the SDA mission, Crider said. “There are so many opportunities for this [AI] technology to augment what we do in space and from space,” she pointed out. “We need AI/ML to help us, not only with the space domain awareness mission, but to help us with decision-making that we need to be able to do, with the optimization of services that we need to be able to provide,” she said. AI/ML tools could also help “optimize the use of imagery and sensing to get the best imagery that we need, when and where we need it. … So there’s a lot of use cases across the board.”