The Looming Swarm

March 22, 2019

DARPA’s Gremlins program, shown in this artist’s concept, envisions launching groups of unmanned aircraft from bombers, transports, and fighters to attack targets while the host aircraft are still out of range. Illustration: DARPA

Swarming technology could find its way to the battlefield within the next few years, at least in a limited capacity, but it will take some time to marry up the artificial intelligence and autonomy needed for a high-end fight.

“I love swarming technology, you probably knew that given the job I came from. I think it’s what future warfare looks like,” said Will Roper, the assistant secretary of the Air Force for acquisition, technology, and logistics.

Roper took over the Air Force’s top technology job in February 2018, after nearly six years at the Defense Department’s Strategic Capabilities Office, where he oversaw development of the Perdix program, among other new technologies.

Perdix are expendable, micro-drones that can be pushed out the back of a variety of military aircraft and fly ahead of larger, more expensive remotely piloted aircraft or manned aircraft to conduct intelligence, surveillance, and reconnaissance missions. The Strategic Capabilities Office, in partnership with Naval Air Systems Command, tested the advanced swarming capability in 2016, launching more than 100 of the micro-drones from three F/A-18 Super Hornets over Naval Air Weapons Station China Lake, Calif.

“Due to the complex nature of combat, Perdix are not preprogrammed synchronized individuals, they are a collective organism, sharing one distributed brain for decision-making and adapting to each other like swarms in nature,” Roper said at the time. “Because every Perdix communicates and collaborates with every other Perdix, the swarm has no leader and can gracefully adapt to drones entering or exiting the team.”

But Roper told reporters in February it was a struggle to even find a range to conduct the test, saying the Defense Department must adapt its thinking to deal with such technology.

“I went to them and said, ‘I’d like to kick 100 micro-UAVs out of fighters,’ and they said, ‘Ok, tell me the flight plan for each one,’?” Roper said. “?‘I don’t have one. They are going to do their own thing, but I can draw a box and make sure they don’t leave that box.’ … We have to shift from ‘you need a flight plan’ to, ‘no you need a box, you need a boundary,’ and that’s OK.”

But before swarming can move “beyond the world of science and technology,” he added, questions do need to be answered: “How do you certify it? How do you test it and evaluate it? Who owns it? Is it a weapons system? Does the platform using it own the autonomy—and swarming and collaboration—or is there a program for swarming and collaboration that plugs that autonomy into all sorts of platforms?”

Air Force test and evaluation, operational testing, and air-worthiness experts will have to get creative to nail down the answers.

Scott Wierzbanowski, a program manager with the Defense Advanced Research Projects Agency’s Tactical Technology Office, said researchers are still learning how swarms might operate and how to incorporate AI algorithms and cooperative autonomy. But he also said swarming, in a basic sense, can still be an effective tool for the warfighter today.

Expendable micro drones.The Perdix micro-drones are capable of low-altitude intelligence, surveillance, and reconnaissance and other short-term missions. They can be air, sea, or ground-launched and operate in both small and large swarms to perform their missions. Source: Strategic Capabilities Office, Perdix fact sheet


Swarming has applications across the full spectrum of warfare. On the high-end, consider an anti-access, area-denial (A2/AD) environment where peer adversaries have deployed multiple integrated air defense systems, making it extremely difficult to penetrate the airspace. Swarms of small, affordable drones could saturate the threat, reducing the risk to manned and more expensive remotely piloted aircraft.

The low-cost drones would share sensor data and work together, Wierzbanowski said, but “if they get shot down, it’s OK.” Losing a drone would be an acceptable cost.

“We really believe that some of these threat areas are going to be extremely dangerous, and we know we need to get close enough to be able to do certain things within those regions.”

Now, consider a clandestine operation in a permissive environment, where a smaller swarm of two to four unmanned aircraft could be launched out the back of a C-130, deploying sensors or widening the path for the C-130—or some other aircraft—to conduct its mission.

In that scenario, “they don’t need the Reapers, or Global Hawks, or the fighters coming in providing that suppression of enemy air defense,” Wierzbanowski said. “They can actually keep it all internal to their own system and be able to do things in maybe a simpler matter, that’s more concise, and in line with what the squadron wants to do. That scalability among different complexities of war is one of the key advantages you get out of this system that you wouldn’t necessarily get out of a legacy-type system.”

Last April, DARPA awarded Dynetics Inc., a 21-month, $38.6 million contract to fund Phase 3 of its Gremlins program. Overseen by Wierzbanowski, Gremlins focuses on the enabling technologies needed to support such disaggregated operations in the future. Specifically, Gremlins sets out to prove that multiple air vehicles can be safely launched and recovered from a C-130 operating far outside of enemy defenses.

After the Gremlins complete their preprogrammed mission, the C-130 would retrieve the air vehicles using a horizontal docking station similar to an air refueling receptacle. Instead of a basket, the dock mechanism would mechanically lock onto the Gremlin to support it, said Tim Keeter, Dynetics chief engineer and deputy program manager for the Gremlins program.

The Gremlins air vehicles themselves are about 14-feet long and weigh about 1,600 pounds when fully fueled. That’s much larger than Roper’s Perdix micro-drones and about the size, Wierzbanowski said, of current cruise missiles.

In Phase 3, DARPA wants Dynetics to recover four Gremlins within 30 minutes in its first big demonstration, slated for January 2020. Longer term, a single C-130 could recover up to 16 of such vehicles, depending on operational requirements. And conceptually, Gremlins could also be launched from F-16s, B-52s, and other aircraft with little modification to the aircraft. That could significantly alter the number of systems in a swarm.

In early February, the team conducted flight tests with its docking station at China Lake, though without an actual Gremlin air vehicle. By early April, Dynetics plans to test the Gremlin’s avionics in a piloted Calspan Learjet.

“It’s like taking the brains of our air vehicle and hosting them on their Learjet,” Keeter said. “It will fly their Learjet around and, of course, if there is any issue or any concerns the human can take over.”

The first Gremlins flight will follow this summer, proving its capability before it is put to work near manned aircraft.

Unlike Roper’s Perdix, Gremlins will not incorporate artificial intelligence or the autonomous behaviors—at least not yet.

Wierzbanowski oversees another program at DARPA that does just that, however. The Collaborative Operations in a Denied Environment, or CODE, program looks at “the autonomy necessary for groups of UAVs to work together, where the operator is more at the strategist level or an oversight level,” he said. “He’s just more providing commander’s intent and the system is able to figure out from that commander’s intent what the system is able to do.”

DARPA tested the CODE UAV’s ability to adapt and respond to unexpected threats in an A2/AD environment late last year at the Yuma Proving Ground in Arizona. The air vehicles initially were able to interact with a supervisory mission commander, but when communications were degraded or denied, the vehicles proved they could accomplish the mission without live human direction.

“The demonstrated behaviors are the building blocks for an autonomous team that can collaborate and adjust to mission requirements and a changing environment,” said Wierzbanowski in a DARPA release.

DARPA will continue to manage the CODE program until this spring, when it will transition to the Naval Air Systems Command. However, Wierzbanowski said if he were “king for a day,” he’d love to bring the Gremlins and CODE programs together to see what could be accomplished.

“Because I have both of them, I’m able to work the advantages of one system on the other,” he explained. “Given that these programs are scoped separately, it’s talking to the services and saying, ‘Hey, listen, … it would be really, really good if we integrated these two programs into a neat demonstration or an experiment where we take the best of breed, put them together, and let’s go see what type of missions we can actually do.’?”

Wierzbanowski said he has weekly conversations with DOD stakeholders and has received some interest, but so far, interest has not produced funding.

Changing on the fly. DARPA’s Collaborative Operations in Denied Environment (CODE) program will enable multiple CODE-equipped unmanned aircraft to collaboratively sense, adapt, and respond to unexpected threats and new targets. The systems can share information, plan and allocate mission objectives, make coordinated tactical decisions, and react in a high-threat environment. Source: DARPA; USAF


Swarming also could be useful in humanitarian operations, like disaster relief. The Air Force Research Laboratory at Wright-Patterson AFB, Ohio, is working with the Wright Brothers Institute and the University of Dayton Research Institute on the “Swarm and Search AI Challenge,” which will run concurrently with a similar competition in the United Kingdom.

The challenge asks participants to figure out how to plan and control simulated swarms of small, unmanned aerial vehicles as they map a wildfire. Teams must use the same UAV design and sensor set, which is based on an AFRL platform, as well as artificial intelligence technology, to come up with the most effective way to accomplish the mission. The UK Forestry Service will post videos for all the teams to outline the scope of the problem, said Mick Hitchcock, senior technology adviser for AFRL small business, in an interview with Air Force Magazine.

“The challenge is focused on a humanitarian mission, but in reality, the learning applies very well to … Air Force interests,” Hitchcock said.

The idea came about last spring when representatives from the UK’s Defence Science and Technology Laboratory visited the Wright Brothers Institute in Ohio. At the time, wildfires were ravaging California, and another wildfire had just caused significant damage in the UK. By making it a humanitarian challenge, the two labs were able to reach out to nontraditional small businesses and universities “who may not want to play on a military mission,” Hitchcock said.

The first scenario is a basic test to get the teams familiar with the software.

The scenarios will get progressively more difficult as the challenge goes on, culminating in a “final showdown” on March 29-31.

Hitchcock said the UK is “doing a lot of work in this arena,” and the US has “already learned a lot from the interaction.”


Roper acknowledged the Air Force “needs to do more with AI,” as well as the challenge posed by matching legacy acquisition rules to develop software and network technology that crosses multiple systems.

To date, maintenance has proven to be “fertile ground” for operationalizing AI, but the stakes get higher with systems affecting missions and lives, Roper said.

Still, researchers are making progress.

“This is no longer something that’s 10 to 15 years down the road,” Wierzbanowski said. “This is something that can be implemented within the next year or two years and actually be used with our current weapons system, or derivatives of our current weapons system.”