U.S. Air Force AI fighter jet draws 30+ test proposals

The U.S. Air Force went looking for three new flight test ideas for its AI-controlled X-62 VISTA research jet and came away with more than 30, after the first research symposium built around the aircraft drew over 80 participants to Edwards Air Force Base, California, in August.

The three-day meeting at the U.S. Air Force Test Pilot School brought together people from industry, academia and the joint force, according to the 412th Test Wing, which released details on Oct. 13. Organizers set out to define three test concepts the X-62 could fly in fiscal 2027 and 2028, but participants submitted a final proposal of more than 30, which the service described as substantial and previously untapped demand for the aircraft.

“We were awed by the level of positive responses and interest,” said Col. Maryann Karlen, commandant of the Air Force Test Pilot School. “This demonstrates, without a doubt, the demand for testbeds like the X-62.”

The X-62, short for Variable In-flight Simulator Test Aircraft, is a heavily modified two-seat F-16D that can mimic the handling of other aircraft and host software that flies the jet while safety pilots sit aboard. It became the first X-plane flown by artificial intelligence in combat-style maneuvering, flying against a crewed F-16 in a 2023 dogfight under the DARPA Air Combat Evolution program.

Briefings at the symposium came from DARPA, the Air Force Research Laboratory, MIT Lincoln Laboratory and the Air Force Institute of Technology, among others. Aircrew from an Air Combat Command fighter squadron also shared recent operational experience on getting pilots to trust machine-learning systems.

The jet is now in the middle of a Mission Systems Upgrade funded by the Test Resource Management Center, which adds an advanced radar, more computing power and room for additional sensors. Lockheed Martin Skunk Works, Calspan Corporation, the Air Force Research Laboratory and RTX are working on the upgrade through the Air Force Test Center. Once finished, the X-62 is meant to work alongside VENOM, the U.S. Air Force’s fleet of F-16s converted for autonomy testing at Eglin Air Force Base, Florida, and to try out new AI uses for high-priority programs.

Ahead of the expected surge in test work, the school flew a series of short missions it named HAVE HOLIDAYS to work out how to integrate and fly new software quickly. They ran in parallel with HAVE HEAT, an April campaign in which an AI agent on the X-62 autonomously intercepted an airborne target using live infrared sensor data.

One of the HOLIDAYS tests, HAVE BALLDROP, put an autonomous agent from a third party outside the usual prime contractors onto the jet for the first time, drawn from an Air Force Research Laboratory program called xFACTOR. Another, HAVE THANKS, flew BirdBox, an open-architecture laboratory software framework that helps sensors recognize targets at the tactical edge.

HAVE DREIDEL tested commercial runtime assurance software from 3Laws Robotics, which acts as a set of guardrails to stop an autonomous aircraft from breaking limits set by its operators. In HAVE SANTA, engineers put IBM’s NorthPole chip, a low-power processor modeled on the way the brain handles information, inside the VISTA’s computers, where it ran Lockheed Martin Skunk Works’ “Supermassive” simulation and calculated parallel flight trajectories in the air.