Questions loom as Air Force robot-wingman effort prepares to take flight
Industry execs outline challenges of technology, test and evaluation—and even philosophy.
Even as the Air Force edges toward a competition to build its long-contemplated robot wingmen, several fundamental questions linger about the autonomous aircraft that will fly into battle alongside manned fighter jets.
The questions are less technological than philosophical, industry executives say. Just how will these “collaborative combat aircraft,” or CCAs, work with manned fighters? How much autonomy should they have? How will their AI be tested?
“A lot of the accomplishments that we've made to date, that the Air Force has made to date, related to autonomy and AI, really demonstrates that level of technology does in fact exist,” said Steve Fendley, who leads the unmanned systems division of Kratos, one of the firms working to develop autonomous drones. Now, Fendley said, policymakers are “deciding where we're going to establish boundaries and how much of that autonomy and AI are we going to allow to be a part of coordinated manned-unmanned operations.”
The Air Force has had several years to contemplate these questions. Today’s CCA effort evolved out of the Avatar program, launched under then-Defense Secretary Ash Carter sometime before 2016. But it took time just to persuade service leaders of the need for autonomous helper drones, said Will Roper, a former Air Force acquisition chief.
This “idea has continued to gain momentum, with more and more Air Force leaders saying this is the obvious thing we need to do in airpower is to break it up so that we get the advantage of distribution and networking,” said Roper, who leads Istari, a digital-design firm.
Service officials now say such drones are needed to provide the “affordable mass” that would help the United States win a war in the Pacific. They want at least a thousand CCAs performing “hundreds of roles,” including flying alongside and out in front of F-35s and their future Next Generation Air Dominance, or NGAD, aircraft.
And they want them more quickly than the typical Air Force acquisition timeline. Secretary Frank Kendall has said he wants CCA production to start before the end of the decade, then reach operational capability on a schedule “comparable” to the NGAD program. The service formally kicked off the CCA program in its 2024 budget request, which asks Congress for $392 million, and officials plan to steadily increase spending over the next five years.
The Air Force will use non-traditional acquisition paths, such as other transaction authority and middle tier of acquisition authority, to move CCAs out quickly, Lt. Gen. James Slife, the service’s deputy chief of staff for operations, said Wednesday during a Defense News conference.
“Our acquisition enterprises are incentivized to move as quickly as possible on this, and so using some of the non-traditional acquisition authorities that we have that allow us to move a little more quickly, to work with multiple vendors and to iterate as we go,” Slife said. “I think those are all part of the strategy.”
But before the Air Force’s vision can become a reality, there are problems the Pentagon needs to overcome and questions industry needs answered.
For one thing, the Defense Department’s testing organizations, while skilled at wringing out hardware, have far less experience in evaluating the kind of cutting-edge AI software that will drive the CCAs.
“How we test new technology, how we have the workforce to understand it, how we have a workforce that's experienced and confident enough that they can get to a ‘yes’ answer in the face of some uncertainty—that concerns me,” said Brett Darcey, vice president and head of product for Shield AI, which makes autonomy systems for drones.
Another problem, Darcey said, is that there aren’t a lot of robot-wingman prototype airframes to develop on. His company recently demonstrated autonomous teaming with a trio of its 9-foot V-BAT drones equipped with its Hivemind autonomy package. But it’s harder to find ways to “pose the complexity challenges that a real-world CCA would pose and allow you to kind of run those engineering steps you need to take,” he said.
Darcey noted that this means industry is building the software and hardware for a new type of weapon concurrently.
“That’s always dangerous,” he said. “Any program where you're trying to build something, you try not to have five or six things changing at one time. You need to have a little more focused time of arrival, and so that's going to be a challenge too.”
One of the toughest hurdles—one that’s shared with many envisioned military robots—is persuading Air Force pilots to trust that their robot wingmen will do what they need to do: fly in formation, respond to inputs, make good decisions, and so on.
“Because ultimately, the goal for all of this, in a manned-unmanned teaming environment, is that the warfighters really can depend on the autonomy that exists within their unmanned teammates,” said Mike Shortsleeve, vice president of strategic development at General Atomics Aeronautical Systems.
Shortsleeve’s company recently won a contract from the Defense Advanced Research Project Agency, or DARPA, to complete the final demonstration and test phase of the LongShot drone, which is built to fire air-to-air weapons while flying beside or ahead of manned fighters.
Kratos’ Fendley says his company’s four-year-old XQ-58 Valkyrie has practiced flying with many Air Force fighters, and while there was initial trepidation from pilots, Fendley said they got comfortable quickly. However, these are structured operations, and the enterprise still needs to practice how teaming will work in conflict, he said.
“Trust level is high for coordinated events. Trust level remains in question, and this is a general thing, not really just about Valkyrie, but I think it's fair to say remains in question when you have unknowns on the other side of the field,” Fendley said.
Setting aside issues of technology and trust, questions about the CCA’s role in operations must be answered before requirements can be set.
For example, said Istari’s Roper, just how sophisticated and expensive should these aircraft be? If the drone is actually expendable, it won’t have the capability to complete the CCA mission, but if the Pentagon moves closer to a traditional fighter, it will be more expensive and much less dispensable, he said.
“I think that's going to be the challenge, because there will always be a voice in that room when someone's trying to find the knee in the curve and say, ‘This is my best bang for the buck,’ who will say, ‘I want more bang.’ There'll be another voice who says, ‘I want less buck’,” he said.
Roper thinks CCAs should be treated “more like reusable weapons than attritable aircraft,” but the Pentagon must still find the right balance.
Industry gears up
At least one thing is certain: companies are lining up to vie for various parts of the CCA program.
Kratos, for example, said it’s currently producing 150 uncrewed aircraft a year, and is upgrading its plants to handle Pentagon bulk buys.
“We've been increasing our capital equipment from that perspective, and revising our production strategies to be able to accommodate a much greater production rate than we've had consistently,” Fendley said.
But despite its years of flying, Valkyrie is not an official program of record with the Air Force. It is a part of the Skyborg program, an effort to develop an autonomy core system to enable future drones to operate with manned aircraft. Fendley said Valkyrie is a “sum of different programs with different specific objectives,” and he remained cagey on whether and when it might become its own program of record.
“I think it will be a function of one of the new programs that's been announced. That's where we'll probably see it. It may be that we transition to an existing program, but I think more likely it will be a new-name program in the very near future,” Fendley said.
In the meantime, Kratos is working to equip various drones—and ultimately the 29-foot, jet-powered Valkyrie—with AI pilots made by Shield AI.
“Our expectation is that we'll be flying within a few months,” Shield AI’s Darcey said. “Our plans include a climbing of the aviation food chain inside of Kratos, just like we do with our V-BAT. We're not going to just show up on the Valkyrie, but we'll be flying in on some of their other drones to get up to the Valkyrie here over the next six months or so.”
Another potential candidate is Boeing’s MQ-28 Ghost Bat, a combat drone originally made for the Royal Australian Air Force and now being tested by the U.S. Air Force. Boeing says Ghost Bat can provide “fighter-like performance” and fly more than 2,300 miles.
The company wouldn’t say what it will officially offer for the CCA program, but the lessons from the Ghost Bat “will be an important part of any formal response that we have,” said Krystle Carr, Boeing’s director of autonomous collaborative platforms.
Boeing has the “right test plans” in place to deliver manned-unmanned teaming when the Pentagon needs it, Carr said. “What I can say is both the government teams and the Boeing teams are ready to meet the timelines that are put out there for when we need to have this capability.”
While it’s still unclear how exactly Ghost Bat is being tested in the U.S., that program will “continue over the next few years,” Carr said. “Our primary customer is the Royal Australian Air Force, and that's who we continue to focus on at the moment.”
General Atomics will also compete for future drone work, Shortsleeve said. LongShot will “certainly be an option for those family of systems that the Air Force and the other services are going to require in that future fight,” he said. The company is also developing Gambit, a suite of drone variants, to position itself for a future CCA contract.
“Soon, we will be testing one of our own capabilities from our family of systems that is going to be significant,” Shortsleeve said.
Air Force leaders have said they plan to award contracts to upwards of 30 vendors, including many non-traditional ones. Those might include defense start-up Anduril, a software maker that this week acquired autonomous drone maker Blue Force Technologies. Anduril plans to equip Blue Force’s in-development Fury with its own Lattice mission-autonomy systems.
While the “newer breed” of autonomy companies are delivering solutions quickly, Darcey said there’s still far to go.
“It's not going to be: is the software mission-ready? It's going to be: have we checked it out enough to make sure that we trust it, to make sure that it works in the hardware environment that it has been targeted to? Those things have to arrive at the same time, and we're still years away there,” Darcey said.
Air Force officials have said they will disclose more details on the CCA program at the Air Force Association’s upcoming Air, Space & Cyber Conference in National Harbor, Maryland.