Skunk Works tests AI in air-to-air combat

(Editor's Note: This post has been updated to reflect additional information received from a Skunk Works spokesperson.)

Skunk Works, a subset of defense prime Lockheed Martin, has “successfully” demonstrated artificial intelligence in air-to-air tactical intercept scenarios, the company said Wednesday.

In partnership with the University of Iowa’s Operator Performance Laboratory, Skunk Works sought to integrate its autonomy software into one of OPL’s L-29 Delfin jets that were produced between 1963 and 1974. The AI-enabled aircraft was then flown in several simulated scenarios to test machine learning algorithms, “from standard head-to-head flights to off-aspect encounters, missile support and missile defense scenarios,” according to a Lockheed Martin news release.

“The team executed simulated-to-real transfer test objectives against a virtual adversary in offensive and defensive risk postures,” the company said, noting it was “encouraged to see clean sim-to-real transfer of learned behaviors and that the AI agent appeared intentional and decisive in its actions.”

The exercises were conducted at the University of Iowa OPL airspace in Iowa City, Iowa on May 22, a Skunk Works spokesperson told Inside Defense. There were two, single-L-29 sorties, they said, and eight test cases were conducted per flight, the news release stated.

“This was the first live exercise of the new flight interface; it's thrilling to see the separate components successfully integrate on the L-29 to demonstrate new capabilities. The complete system performed even better in live flight than in simulation," Tom Schnell, OPL professor at Iowa Technology Institute, said in a statement.

The demonstration comes as the Air Force for ways to quickly and affordably develop and field innovative capabilities crucial for a future fight with a near peer adversary like China or Russia.

As part of the service’s recently announced “Reoptimization for Great Power Competition” concept, the Air Force is trying to change the way it solicits weapon systems, from a long-term sustainment structure to the more iterative and agile model planned for the Collaborative Combat Aircraft, or drone wingmen meant to tag along manned fifth- and sixth-generation fighters.

“We want to incentivize industry in a way that will allow us to move at the speed of technological advancements,” Air Force Chief of Staff Gen. David Allvin said in May. If the Pentagon can’t take advantage of “fleeting and frequent” technological advances due to the terms of a contract, “that’s where you fall behind as an institution,” he added.

Additional flight tests under the advanced automation program are planned for the remainder of this year, Skunk Works said, in which the team will create more complex scenarios by introducing several aircraft into counter air and battle management situations.

"We’ve developed our AI with an open systems architecture, enabling us to deploy to compatible platforms," the Skunk Works spokesperson said. "This test evaluated AI in isolation against a single, virtual adversary. Additional flight tests this year will increase in complexity, culminating in a collaborative multi-ship crewed-uncrewed teaming air-to-air scenario."

Last month, Air Force Secretary Frank Kendall flew on board an AI-enabled X-62A VISTA in part using Skunk Works software. The modified F-16 Fighting Falcon engaged in a high-intensity dogfight with a manned F-16, allowing the platform’s algorithm to perform tasks, react to new conditions and make decisions based on safety and weapons effectiveness -- all while Kendall watched.

"Our work on the X62-A VISTA for the U.S. Air Force Test Pilot school is an example of the hardware modification required for transfer to an existing system," the Skunk Works spokesperson added.