DARPA wants to automate aerial dogfighting with AI wingman

DARPA has launched an initiative to develop an artificial intelligence (AI) that can engage in dogfighting aerial combat. Despite its Terminator-like description, the purpose isn't to create some sort of Skynet cybernetic fighter ace, but rather to produce an AI wingman that human pilots can trust to watch their backs while they handle command decisions.

Artificial intelligence has made great strides in recent decades and now DARPA wants to install an AI in unmanned or man-optional aircraft that is so sophisticated that it can handle one of the hardest of fighter pilot tasks – aerial combat. In some ways, this challenge is pushing the current envelope of what AI can accomplish, though in others, dogfighting is well suited to the technology's strengths.

According to DARPA, the current state-of-the-art is still on the human's side because a flesh and blood pilot is still streets ahead when it comes to high speed, high-G, seat-of-the-pants, split-second fighting. However, if an AI can master the art, then the computer's speed and memory would be an invaluable asset that would free the pilot to concentrate on the bigger picture involving control of multiple aircraft and missiles. The trick is to come up with an AI that is actually up to the job.

AI has already proven itself very good at handling situations where the rules are clear and it's easy to evaluate outcomes, which is it does so well at Chess, Go, poker, and Jeopardy. Dogfights may be highly nonlinear in nature, but their outcomes are objective and how the aircraft operates is strictly limited by the laws aerodynamics – giving the AI the limited, predictable world that it needs to excel.

So, DARPA has initiated the Air Combat Evolution (ACE) program, which kicks off with a Proposers Day on May 17, 2019 in Arlington, WA. It's one of several programs that the agency is running to promote what it called "mosaic warfare." That is, producing systems that rely much less on the human aircrews and more on a blend of human and machine intelligences that build on the strengths of each. In addition, such mosaic systems can be very quickly developed and upgraded as technology progresses, as well as being highly adaptable to new situations by replacing elements as needed.

In its present form the ACE program will teach AI how to dogfight in a similar way to that of human trainees. It will be taught the basic rules, then progress to simple one-on-one scenarios before going on to more difficult missions. As it does so, fighter instructor pilots seated inside the autonomous aircraft will evaluate its prowess and help them to develop tactics.

DARPA says that, because dogfighting is very rare in these days of over-the-horizon engagements, the point of the program isn't to make a robotic dogfighter. Instead, the purpose is the same as for dogfighting training for human pilots, which is to place them in high-pressure situations that allow performance and trust to be refined as pilots advance to become mission commanders on teams that need to rely on one another. It's this reliability and trust that is the true goal of ACE.

"Being able to trust autonomy is critical as we move toward a future of warfare involving manned platforms fighting alongside unmanned systems," says Air Force Lieutenant Colonel Dan Javorsek (PhD), ACE program manager in DARPA's Strategic Technology Office (STO). "We envision a future in which AI handles the split-second maneuvering during within-visual-range dogfights, keeping pilots safer and more effective as they orchestrate large numbers of unmanned systems into a web of overwhelming combat effects.

"Only after human pilots are confident that the AI algorithms are trustworthy in handling bounded, transparent and predictable behaviors will the aerial engagement scenarios increase in difficulty and realism. Following virtual testing, we plan to demonstrate the dogfighting algorithms on sub-scale aircraft leading ultimately to live, full-scale manned-unmanned team dogfighting with operationally representative aircraft."

By the time ACE reaches Phase 1, DARPA will sponsor a series of "AlphaDogfight Trials," which will involve automating individual tactical behavior for one-on-one dogfights by pitting competing algorithms against one another in a tournament-style competition.

Source: DARPA