The U.S. Air Force’s test enterprise has been at the vanguard of some of the most consequential technological advancements in aviation’s history. From Chuck Yeager breaking the sound barrier in 1947 and the first landings of the space shuttle in 1981, to today’s testing of Artificial Intelligence in manned and unmanned aircraft, America’s Air Force has pushed the limits of human endurance and innovation for decades.

  While AI and autonomy might appear as recent phenomena, bolstered by the accessibility of Large Language Models like ChatGPT, Google Gemini, and others, their roots in aviation stretch back over a century. “Autonomy in the rules-based form has been around since really the dawn of aviation,” said Col. James “Fangs” Valpiani, U.S. Air Force Test Pilot School commandant.

  “The first automated aircraft, the first autopilots were implemented in the early 1910s, so right after the Wright Brothers. You can accomplish an incredible amount with that kind of autonomy, in fact, the Air Force has all manner of automated drones as well as automated space probes that use this type of coding.” 

  Recent developments have markedly increased the scope of AI, enhancing tasks as routine as paperwork processing and as complex as the advancement of kinetic warfare technologies. The Air Force Test Center is at the forefront of the Air Force's exploration into AI and autonomy and leverages its vast institutional knowledge, resources and the skilled ingenuity and creativity of its testing experts to craft innovative solutions that address the challenges confronting the Air Force today and into the future.

Col. James “FANGS” Valpiani, United States Air Force Test Pilot School commandant, poses for a photo at Edwards Air Force Base, Calif., March 6, 2024. The school is where the Air Force's top pilots, navigators and engineers learn how to conduct flight tests and generate the data needed to carry out test missions. In addition to pilots, they also train weapons systems officers, combat systems officers and Guardians. (U.S. Air Force photo by Tech. Sgt. Janiqua P. Robinson)

  Originally the Air Force Flight Test Center, AFTC was redesignated in 2012 and is headquartered at Edwards Air Force Base, California. It manages the Air Force's comprehensive flight test operations, including three primary test bases: Edwards AFB, Eglin AFB, Florida and Arnold AFB, Tennessee. Maj. Gen. Evan C. Dertien, AFTC commander, emphasized the center's critical role in propelling the development of autonomous aircraft technologies. 

  The test enterprise’s culture of innovation and history of technological advancements plays a crucial role in shaping the integration of AI within the Air Force. Dertien explained, "We possess the necessary safety protocols, testing infrastructure, data and risk management capabilities. So, it's only logical as the Air Force starts to adapt across both warfighting and personnel platforms that the test center would be at the forefront. We're excited to be the caretakers of evolving that technology.”

  The Test Pilot School (TPS) and the X-62 Variable Stability In-flight Simulator Test Aircraft (VISTA) are leading the way in AI and autonomy testing. This platform enables flight test professionals at TPS to delve into the optimal application of AI in kinetic warfare platforms. Initially developed as a custom F-16 test bed for exploring thrust vectoring — a series of tests that laid the groundwork for the F-22's thrust-vectoring capabilities — the X-62 (previously known as the NF-16) has been adapted for various purposes over the years. “What makes the X-62 unique is that it can mimic the dynamics and handling characteristics of a diverse array of aircraft, from a C-17 to a 747, and even a Cessna," Valpiani noted. This versatility renders it an exceptional resource for training TPS students on numerous aircraft types. "We recognized that by virtue of being able to simulate a wide range of flight dynamics, the X-62 had the potential, if we could layer on autonomy, to become a game-changing platform once again. We've collaborated with the Air Force Research Laboratory (AFRL), the Defense Advanced Research Projects Agency (DARPA) and AFTC in recent years to develop what we call an 'autonomy sandbox.'"

  The X-62 is designed for experimenting with autonomy based on machine learning, diverging from the more traditional rules-based programming historically employed in autonomous systems. There are challenges faced with this type of coding, however, namely, its necessity for a predictable environment, where human experts can code clear, universally applicable rules.

The X-62 VISTA receives post-flight maintenance after a flight at Edwards Air Force Base, Calif., March 7, 2024. VISTA stands for Variable Stability In-flight Simulator Test Aircraft and is operated by the Air Force Test Pilot School with the support of Calspan and Lockheed Martin. (U.S. Air Force photo by Tech. Sgt. Janiqua P. Robinson)

  “If you think about the kinds of challenges that the Air Force and aviation in general faces going forward, we’re talking about very dynamic environments,” Valpiani explained. “To tie this type of autonomy into everyday life, consider the cruise control function on your car, it's a very basic system, it’s expert coded, you tell it what to do … if you set it to 30 mph, it’ll hold that setting until you tell it to stop,” Valpiani said. “But if you were to try to generate a set of code and rules for how the vehicle should drive through say, downtown New York City traffic, you can imagine how a dense urban environment — with pedestrians jay-walking, bicycles out of their designated lanes, etc. —  is an extraordinarily dynamic situation.” Machine-learning algorithms help to solve the complex problems presented by dynamic situations by using statistical weightings and mathematical formulas to derive patterns that humans wouldn’t necessarily be able to see on their own.

​  “Machine learning has great promise for Air Force aviation, but the nature of machine-learning algorithms is that it’s hard for humans to scrutinize and understand these algorithms, we can ask it to do the same thing twice, and potentially get two completely different outcomes,” Valpiani said. Solving this unpredictability element is particularly important for military applications, especially combat autonomy, where understanding and predicting behavior is crucial. “As a result, we need a more robust safety environment with which to test and develop machine learning before we can be certain that they’re usable in less controlled environments,” Valpiani explained.

The X-62 Variable In-Flight Simulator Test Aircraft (VISTA) flies in the skies over Edwards Air Force Base, California, March 23, 2023. (U.S. Air Force photo by Ethan Wagner)

  The researchers at TPS don’t just test these platforms for efficacy. One of the key aspects of testing is ensuring that the machine learning algorithms used in the VISTA (and other platforms) conform to the same rules of responsible use that the Air Force expects of all human pilots, “so in our partnerships with DARPA and other agencies, our main focus has been ensuring that their algorithms developed in simulated environments comply with those rules and norms of responsible use,” Vapliani stressed. “Then when we bring them over into real-world application, we’re assessing the same compliance and responsibility characteristics to ensure that they behave according to all the same rules that we hold ourselves to as human pilots.”

  “You really can’t talk about AI without considering the ethical and responsible use of the technology. Everything we’re doing right now from the grassroots forward is to make sure we comply with the DOD’s five principles for the responsible development of AI.”
Maj. Gen. Evan C. Dertien, Commander, Air Force Test Center.

  The principles guiding these tests are applied Air Force-wide, irrespective of the platform being tested. “You really can’t talk about AI without considering the ethical and responsible use of the technology,” Dertien emphasized. “Ultimately everything we’re doing right now is trying to develop and mature AI responsibly and ethically. The DoD has its five principles for the ethical development of AI: AI must be responsible, repeatable, traceable, governable and equitable, and everything we’re doing right now from the grassroots forward is to make sure we comply with those principles.”