Via Satellite

Satellite Leaders Embrace Machine Learning Revolution

Machine Learning (ML) is an industry game changer, with the potential to impact the way satellites are operated and automated. But while we’re accustomed to our home-based Artificial Intelligence (AI) devices, we’re only beginning to scratch the surface of ML and AI in the satellite industry, said speakers at the SATELLITE Show’s packed Monday afternoon session, “How Machine Learning Offers to Revolutionize Satellite Operations.”

“Twenty years ago, AI was very new, very exciting,” said speaker Rajeev Gopal, Vice President (VP) of advanced systems at Hughes. “With AI and ML, there’s a great bit you can automate.”

Spencer Ziegler, director of space U.K. at SCISYS, said ML applications can be leveraged for infinite purposes, like improving the Mars rover — which can use AI-fueled observations to figure out where to travel and what data to collect on the Red Planet. “[ML] helps us explore the lifecycle end-to-end of how a space mission works,” said Ziegler. “With the mission to Mars, we’re integrating aspects of ML to the Rover to help identify objects and record data.” In past GPL missions the Rovers did not have autonomy, and often missed important data, he added.

For Maria Demaree, VP and general manager of space mission solutions at Lockheed Martin, advances in ML have fulfilled personal — as well as professional — aspirations. “It took me 30 years to get to this panel,” said Demaree, recalling her college years, when she dreamed of working with AI. “It’s finally coming to the point where we’re embracing it. When we were trying to do AI 30 years ago, we were just trying to build the computer to be as smart as it could be. The ability to think past those four corners wasn’t there. I would say that over the past three to five years it’s become very exciting and now you don’t have a conversation where it doesn’t come up.”

Lockheed Martin is currently focusing its research on areas such as image recognition and natural language processing — specifically, how ML capabilities can enhance researchers’ and scientists’ jobs, she said.

But when session moderator Caleb Henry, staff writer at Space News, asked whether AI would soon automate most satellite industry workers’ jobs, the answer among all three speakers was a clear “no.”

“We’re really not trying to turn over everything to robots,” Demaree said. “It’s about providing our analysts, our users, a way to be more accurate and more efficient.”

Ziegler said there’s often a perception in consumer or trade media that one simply can “plug something in” and start reaping profits. “That’s not how it works,” said Ziegler. “In satellite operations … you’ll be using ML to support humans. It will let us work with more systems, more complex systems.”

Gopal agreed. In December, the service provider made headlines for its successful demonstration of a new AI-based enterprise management and control technology for military leaders. The technology — formally known as the Flexible Modem Interface (FMI) — was designed as part of a U.S. Air Force program, and uses an AI rules engine to derive insights. “If you look at today’s systems …some are only 80, 90 percent accurate,” he said. “You can’t replace humans.”

Demaree said that many, if not most, jobs require close collaboration between dedicated workers and the back-end algorithms that produce insights. “When we look at imagery and change detection, that’s a manual process,” said Demaree. “If we could have a computer to say, ‘this is a naval yard, this is a machine yard’ … you really expand the capability of our analysts. We’re giving more efficient capability to end users. Our approach isn’t to eliminate the human in the loop.”

Speakers also discussed the importance of leveraging ML for emerging applications, such as improving satellite-system monitoring by alerting end users if performance deviates from established boundaries, or thresholds. In time, more satellites will be equipped with self-monitoring capabilities, which will alert operators to problems quickly.

Yet there are still some challenges around data — from gathering the right information at the right time, to sharing information with other providers. Henry pointed out that because the space industry is smaller than other industries, such as telecom, there are concerns over whether there is enough data available to drive decisions. There are also concerns that too much data exists in silos, which could stymie progress.

“I do believe there is the right amount of data, but the question is can we get to it?” said Demaree, suggesting there are security issues that put restrictions on providers sharing all data all the time. “How do we bring the data sets together in the cloud? That’s the challenge.”

Another challenge is figuring out how to sift through the deluge of unimportant data to get to the mission-critical insights. “There’s a lot of data being generated, whether it is Internet of Things (IoT) or Earth Observation (EO),” said Spencer. “But a lot of it is of no use to us.” VS