Lockheed Martin’s Skunk Works unit and pilots from the University of Iowa’s Operator Performance Laboratory have successfully tested the use of artificial intelligence agents in carrying out an electronic attack task.
Represented by two piloted L-29 aircraft, the AI agents were able to provide jamming support in a simulated air-to-ground mission, Lockheed Martin announced from Iowa City, Iowa on Wednesday.
Joe Villani, vice president of Lockheed Martin’s 21st Century Security Demonstrations & Prototypes unit, said the initiative is a “terrific example” of applying AI and unmanned aerial systems to realize the Department of Defense’s Joint All Domain Operations vision.
During the program, known as Enhanced Collaborative High-Frequency Orientation System, the L-29 aircraft were flown by OPL pilots who followed directions in the form of heading, altitude and speed cues from the AI agents.
Trained with advanced deep multi-agent reinforcement learning methods, the AI agents became capable of using emergent collaborative tactics with minimal reward shaping and an emphasis on mission completion. The assessment centered sim-to-real transfer to train the AI agents in simulated environments and carry that understanding into use on real aircraft.
On the first day of the test, the agents aligned between the fighter and target within a tight tolerance, preventing the radar from getting a track on the fighter. Over the three-day period, the team was able to complete every test card ahead of schedule.
Tom Schnell, founder and director of the OPL, said the test was an “exceptional example of industry and academia working together on a cutting-edge project.”
Skunk Works and OPL aim to evaluate AI in an end-to-end Suppression and Destruction of Enemy Air Defenses mission prior to the end of 2023. Information gathered from these tests is intended to drive future AI and automation development for upcoming programs, including the U.S. Air Force’s upcoming Collaborative Combat Aircraft, and current crewed-uncrewed teaming development.
