On May 2, Secretary of the Air Force Frank Kendall took a front-row seat in aviation history, flying in the front seat of the X-62A VISTA at Edwards Air Force Base. The one-of-a-kind aircraft is a modified F-16 fighter jet fitted with advanced machine learning and specialized software that are being used to test new autonomous flying capabilities among other advanced technologies.
The X-62A, better known as VISTA or Variable In-flight Simulation Test Aircraft, is singular in its mission of real-time development and testing of flying capabilities with machine learning and live agent integration. Partly developed under the Defense Advanced Research Projects Agency’s ACE or Air Combat Evolution, program, this technology has been a finalist for the 2023 Robert J. Collier Trophy for its breakthrough achievements.
Kendall waxed poetic on the transformational possibilities of autonomous air-to-air combat: Theoretically, this concept has been possible for many decades, but it’s possible now. “In 2023, the X-62A broke one of the most significant barriers in combat aviation,” he boasted, bragging about the accomplishment of the ACE team.
The U.S. Air Force Test Pilot School’s research division leads overall program management for the X-62A and concentrates on accelerating multidomain warfighting capability through rapid testing of novel technologies. The team, led by the research division, began its nearly four-year process to upgrade VISTA’s test-training capability used to simulate other aircraft flying characteristics. They conceptualized this development to mean the use of VISTA as a tool upon which the theory in artificial intelligence could be integrated and applied using live agents in real-time. This dream was fulfilled when, in 2022, the upgrade featured three new highly specialized software suites with significantly higher computing power.
AI takes the best available technology and automates the decisions that previously had been made by humans,” Kendall said. On this flight, the X-62A conducted a range of tactical maneuvers with live, responsive agents reacting in real time to simulated threats. Neither Kendall nor the safety pilot strapped into the backseat ever touched the controls during this test flight, which helped validate the models and test the aircraft’s performance.
The flights were part of a series of Air Force events focused on developing, testing, and maturing those autonomous capabilities as part of service platforms. The Kendall demonstration flight was an important first step in translating AI-enabled operations from computer simulations to real-world environments.
During the exercise, an autonomous agent developed by Shield AI was onboard the X-62A VISTA, engaging a manned F-16 in a within-visual-range air combat scenario dogfight. Such is Shield AI’s reinforcement learning-based AI, a descendant of the technology to won DARPA’s AlphaDogfight Trials in 2020.
The idea is you want to train algorithms to continue to perform desired outcomes through trial and error, like how humans make decisions,” said Brett Darcey, executive vice president at Shield AI. The aim is to reward desired behaviors and penalize undesired ones so the algorithm learns what to do by interpreting its environment.
Through most of Kendall’s flight, VISTA did high-intensity demonstrations of the capability of an AI agent to control the aircraft and maneuver into a position where it could engage with the manned F-16. Kendall initiated automation by pressing a button on his stick to permit the AI to take control in a series of combat scenarios.
Darcey said the AI had been specifically designed for dogfighting, but the modular building blocks of its architecture could be taken out and replaced for other missions. That flexibility will enable it to perform even more complicated tasks, like engaging in combat and refueling mid-flight.
Shield AI is working, too, via a partnership with Kratos to let pilots control AI agents in real time and update their models to adapt to new conditions. This forms “plannable, directable, and trustable,” autonomy, Darcey said.
Looking ahead, Shield AI says it will continue maturing its AI pilot through work with DARPA and a partnership with Kratos while expecting future programs focused on bringing autonomy to the Air Force’s manned and unmanned platforms. Meanwhile, the Air Force is moving ahead with its program for collaborative combat aircraft, the service’s effort to develop loyal wingman drones to fly alongside manned aircraft by the end of the decade.
Kendall imagined a future in which AI agents would fly in combat and outcompete human pilots. He mentioned some advantages of AI, including the fact that computers don’t get fatigued or terrified and can process many information inputs. But he added that the laws of armed conflict must be followed and that automated weapons must avoid unnecessary destruction of property and injuries.
But more importantly, with more advanced AI-enabled drones, combined with other autonomous technologies, the Air Force hopes to integrate all of these by 2030 into the force as part of the revolution of the future of combat aviation.