Flight tests overseen by the Defense Advanced Research Projects Agency and the Air Force showed the AI-enabled autonomous fighter jet to be safe and secure, including “nose-to-nose” dogfights with human F-16 pilots, officials said. It has been proven that it can be used effectively.
A few years later, during DARPA's AlphaDogFight trials, the algorithm went undefeated in a simulated computer battle with military aviators. Most recently, the agency's Air Combat Evolution program used a modified F-16 known as the X-62A VISTA (Variable Flight Simulator Test Aircraft) to put machine learning agents through their paces in the skies above Edwards Air Force Base. Masu. California.
A total of 21 test flights were conducted for the project between December 2022 and September 2023, the Pentagon said in an ACE program update released Wednesday.
“Starting in December 2022, this marks the first application of machine learning agents to control the flight paths of fighter jets,” Col. James Valpiana, Air Force Test Pilot School commander, said in a video accompanying the update.
More than 100,000 lines of flight-critical software changes were made to improve the tool.
And in September, “We actually took the X-62 and flew it against a manned F-16 in combat, first for defense, then for attack, then at 1,200 miles per hour and up to 2,000 feet.'' We achieved safety through a high-aspect nose-to-plane engagement,” said Lt. Col. Marian Karlen, deputy commander. Test Pilot School said in a video.
The Pentagon said in a program update that the exercise was “the first AI-on-human visual range engagement (also known as a 'dogfight') conducted using actual manned F-16 aircraft.” Stated.
At the time of publication, the department had not provided information to DefenseScoop about whether the machine learning agent controlling the X-62A defeated the human pilot in that encounter.
But defense officials emphasized the importance of these efforts to demonstrate that artificial intelligence technology can operate safely in complex combat environments such as close quarters combat and air-to-air combat.
“Prior to formal verification methods for AI-based autonomy, the team will train AI agents to comply with safety requirements such as flight envelope protection and air/ground collision avoidance, as well as ethical requirements such as combat training regulations. and pioneered new ways to test , weapons engagement zones, and intercept paths of fire,” according to a Department of Defense program update.
“The X-62A team has demonstrated that state-of-the-art machine learning-based autonomy can be safely used to fly dynamic combat maneuvers. We accomplished this while complying,” Air Force Secretary Frank Kendall said in a video, adding that the capability was “transformative.”
Officials said the autonomous aircraft had a human pilot on board in case something went wrong during the test flight and a handover was needed, but that It was noted that there was no need for staff to activate the safety switch.
Other organizations supporting this program include Calspan, Cubic Corporation, EpiSci, Lockheed Martin Skunk Works, physicsAI, Shield AI, University of Iowa Operator Performance Institute, Johns Hopkins Applied Physics Laboratory, and MIT Computer Science and Artificial Intelligence Laboratory. , MIT Lincoln University, and others. Laboratory.
Kendall, a strong supporter of the U.S. military's use of AI, told lawmakers last week that he intends to fly an F-16 in autonomous flight mode later this year.
Kendall said the success of the ACE program contributed to the decision to move forward with the Air Force's Cooperative Combat Aircraft (CCA) program, which develops and operates next-generation autonomous unmanned aircraft for anti-aircraft operations and other missions. The service plans to spend billions of dollars on this effort over the next few years.
“The critical issue on the battlefield is time. And AI will be able to do much more complex things, much more accurately, and much faster than humans. If humans get involved, we lose. It can also be monitored and the behavior of the AI can be monitored. [but] If you try to intervene, you're going to lose,'' Kendall said during a panel discussion at the Reagan National Defense Forum in December.
“I just received a briefing from DARPA about some research they're doing on manned-unmanned combat, basically fighter aircraft,” Kendall said. “AI always wins, depending on how you structure your tests. But the difference between how long it takes a human to do something and how long it takes AI to do something makes a big difference in results. We're talking seconds here. To give you a sense of the parameters, the best pilots you'll find so far will take tenths of a second to do something. does it in microseconds, which increases performance by orders of magnitude. And that time actually matters. And there's no getting around that. But that's the reality we have to face. It is.”
