USAF Plans For Its "Skyborg" AI Computer Brain To Be Flying Drones In The Next Two Years

The service envisions the artificial intelligence system eventually flying larger, more complex unmanned designs and in the cockpit with human pilots.


The U.S. Air Force is working on an artificial intelligence "computer brain," dubbed Skyborg, that it hopes will be able to autonomously operate a prototype “loyal wingman” drone, such as the Kratos XQ-58A, by 2021 and be able to fly an actual “fighter-like aircraft” two years after that. The larger goal is to integrate this system into various unmanned planes in the future and eventually turn it into a digital co-pilot to help human aviators in manned platforms.

On Mar. 20, 2019, Aviation Week was first to report the projected timeline for developing Skyborg and the first prototype aircraft equipped with the system. The Air Force Research Laboratory had issued a formal request for information to determine what technologies might be available already, or close to available, to support Skyborg, on Mar. 15, 2019. Two days before that, Will Roper, Assistant Secretary of the Air Force for Acquisition, Technology and Logistics, had publicly disclosed the program for the first time at a conference consulting firm McAleese and Associates and bank Credit Suisse had sponsored in Washington, D.C.

“Skyborg will directly support the initiatives outlined in the 2018 USAF Artificial Intelligence Strategy and the 2019 Executive Order of Maintaining American Leadership in Artificial Intelligence (AI) while filling an immediate operational need,” the Air Force Research Laboratory’s contracting notice explained. “The primary goal of the Skyborg program is to deploy a modular, fighter-like aircraft that can be used to quickly update and field iteratively more complex autonomy to support the warfighter.”

First and foremost, Skyborg is about developing an AI system that can fly an unmanned aircraft with minimal, if any human interaction. The Air Force is also adamant that they want it to use a software that is “modular” and “open architecture” so that it can rapidly add new “complex autonomous behaviors” into this advanced computer brain in the future.

Kratos' XQ-58A, one drone that could eventually fly with the Skyborg AI, during its first test flight.

“I expect the first things that we'll do will not appear as sexy as what you might imagine in a movie, but will be completely game-changing,” Roper explained at the conference on Mar. 13, 2019. He did not elaborate, but one of the immediate benefits of an AI or AI-enabled system is the increased speed of decision-making.

At first, a drone with Skyborg might not necessarily do anything a manned aircraft can’t do already, but it will be able to perform those tasks, such as aerial combat maneuvers or weapons employment, faster based on its set parameters. Where a human might be distracted or confused by the chaos of an aerial engagement, an autonomous unmanned aircraft would simply act.

The drone would be able to make its decision quickly, but also based on an immense amount of situational data that would take an actual pilot much longer to process. This alone is a major leap in capability and would serve as a jumping off point for the development of more advanced capabilities going forward, which you can read about in more depth here.

The Air Force also has a “critical” requirement for at the least initial prototype drone design to be “attritable.” The service is keen to make clear that this term doesn’t mean “expendable,” but does describe a design that is low cost enough for commanders to more readily consider sacrificing it on the battlefield if necessary.


The Air Force's publicly announced critical requirements for the Skyborg program as of March 2019.

If the drone could already have a terrain avoidance and all-weather flight capability, be able to take off and land on autonomously, and feature modular payload bays for sensors and other systems, the Air Force Research Laboratory says this is “highly-desired,” though not essential, at least at the beginning. Lastly, the Air Force says it is “desired” that the Skyborg systems and associated aircraft be still be easy to use even for personnel who have limited or no engineering or pilot background if at all possible.

These latter requirements seem very focused on the Air Force’s desire to rapidly transition Skyborg from an experiment into an operational capability. This is something that Assistant Secretary Roper stressed in his earlier remarks.


Non-critical requirements for Skyborg.

“I don’t want this to just be a laboratory project that lives and dies there in a petri dish. I want this to become a program,” he explained. “I want to see real, operational demonstrations within a couple years. And I will push them to be faster than that.”

This also strongly suggests the Air Force will be inclined to use a platform already in development, or even operational, to host the first iterations of Skyborg. Roper did not say whether the service had a specific platform it was focusing on, but did mention the possibility of putting the AI into target drones such as the QF-16 or BQM-167.

In 2016,  drone maker Kratos demonstrated its Unmanned Tactical Aerial Platform-22 (UTAP-22), also known as the Mako, an evolution of the BQM-167,  in the loyal wingman role together with a manned U.S. Marine Corps AV-8B Harrier jump jet. The Air Force has also previously employed modified, unmanned F-16s in tests of the loyal wingman concept, known as Have Raider and Have Raider II, in 2015 and 2017 respectively. All of these experiments could help further inform the Skyborg program's requirements. 

He also mentioned the possibility of pairing Skyborg with Kratos’ XQ-58A Valkyrie, which just flew for the first time earlier this month, as seen in the video at the beginning of this piece. The design already meets the critical attritable requirement and AFRL is already working on it as part of its Low Cost Attritable Aircraft Technology (LCAAT) program, which would make it attractive starting place.

The Air Force is interested in the XQ-58A and LCAAT as stepping stone to an operational and cost-effective surveillance, strike, and electronic warfare platform. These drones could operate singularly, as part of a networked, autonomous swarm, or in groups of loyal wingmen in cooperation with manned aircraft, revolutionizing how the Air Force operates.


An overview of the Air Force's LCAAT program.

An AI system such as Skyborg that has the ability to grow and evolve over time is almost a requirement for the XQ-58A, and any future “fighter-like” unmanned combat air vehicle (UCAV) designs, to carry out any of these mission sets most effectively as the new threats and challenges emerge. There may also be a benefit to integrating the system into manned aircraft, offering a sort of virtual co-pilot to reduce the burdens on human aviators.

“I might eventually decide, 'I want that AI in my own cockpit,'” Assistant Secretary Roper said at the Washington, D.C. conference. “So, if something happened immediately, [Skyborg] could take hold, make choices in a way that [a pilot would] know because [a pilot has] trained with it.”

Again, the benefit that Skyborg would bring the table first would be its ability to make those decisions faster and incorporate amounts of information that would overwhelm a real pilot to help make the best choice possible. It could also simply spit back a menu of possible options for an aviator to rapidly choose between.

USAF capture

In March 2018, the Air Force Research Laboratory announced the Air Force 2030 initiative and released a video presentation highlighting various technologies the service might implement in the future. It included this sequence where a pilot in an F-35 Joint Strike Fighter appears to engage with a virtual co-pilot with working together with loyal wingmen drones.

It could potentially provide revolutionary improvements in survivability and safety. Skyborg could help manned aircraft avoid or evade enemy defenses and taking over for a pilot suffering from hypoxia or wounds from enemy fire. Tapped into an aircraft’s own computerized diagnostics systems, it could immediately identify faults and either look to correct them or alert the pilot to the issue. The importance of speed of response in in-flight emergencies has already led to the development of the science of what is known as "crew resource management," which involves methods of streamlining communication and troubleshooting at critical moments.

There’s a major value in just being able to shunt a host of relatively minor, routine tasks in flight to a robust AI-driven system, too. This would help reduce fatigue, which, in turn, could lessen the chance of tired aviators making simple, but serious mistakes. The U.S. Army, in cooperation with the Defense Advanced Research Projects Agency (DARPA), is already pursuing a similar effort, known as the Aircrew Labor In-Cockpit Automation System (ALIAS), to develop autonomous flight systems for its helicopters. In February 2019, the Pentagon kicked off a U.S. military-wide push to further integrate AI and machine learning into a host of operational and other activities.


The Army and DARPA are already experimenting with this modified Sikorsky S-76B helicopter, which features a semi-autonomous "co-pilot." 

The Air Force will spend much of Skyborg’s initial phases figuring out exactly what the first generation of the AI can and can’t do and then building out from there. “Before operational AI innovation can occur, the USAF must field an autonomous system that meets an immediate operational need and can serve as an iterative platform to jumpstart complex AI development, prototyping, experimentation, and fielding,” AFRL’s contracting announcement notes.

The hope is that the rapid prototyping and development of actual flying drones with the system on board in the next few years will speed up the process. “I think once we have it out in the operators’ hands, it opens up new opportunities,” Assistant Secretary Roper said.

But this also raises some curious questions about the actual state of the technology the Air Force might be looking at for this program. The service's prior work on UCAVs, at least publicly, evaporated almost a decade ago, despite the service and its industry partners having made major progress on promising technologies that seem very much in line with Skyborg's general goals. By 2003, Boeing's X-45A UCAV demonstrator was already capable of autonomous operation thanks to its Decision Mission-Control Software, or DICE. Lockheed Martin's Skunk Works advanced project division has also been active in developing autonomous unmanned aircraft and swarming capabilities. The War Zone covered all this and much more in a previous in-depth feature.

While Skyborg is certainly an important development, Roper and AFRL have framed it as new. At the same time, one has to wonder how a comparable system doesn't already exist in a classified realm after all these years. The alternative is that the Air Force truly abandoned its previous work entirely and is starting again more or less from scratch. It's a strange situation that The War Zone will be looking to address at great length in the near future.

With the Air Force’s aggressive schedule for Skyborg, we will hopefully be getting more details on what the service has in mind for this game-changing technology in the near future, as well.

Contact the author: