One of the U.S. Air Force's stealthy XQ-58A Valkyrie drones recently completed a successful test flight demonstrating the ability to carry out aerial combat tasks autonomously using new artificial intelligence-driven software. The service says the test is part of a tiered approach to maturing autonomy "agents," which involves training algorithms millions of times first in simulations and other testing. This includes the Collaborative Combat Aircraft program, or CCA, a key part of the larger Next Generation Air Dominance modernization initiative.
The milestone XQ-58A test took place at Eglin Air Force Base in Florida on July 25, according to a press release from the Air Force Research Laboratory (AFRL). The flight lasted approximately three hours and was conducted within the bounds of the Eglin Test and Training Complex, a group of offshore training ranges in the Gulf of Mexico. AFRL's Autonomous Air Combat Operations (AACO) program team developed the algorithms used for the test.
“The mission proved out a multi-layer safety framework on an AI/ML [artificial intelligence/machine learning]-flown uncrewed aircraft and demonstrated an AI/ML agent solving a tactically relevant 'challenge problem' during airborne operations,” Air Force Col. Tucker Hamilton, the service's head of AI Test and Operations, said in a statement. “This sortie officially enables the ability to develop AI/ML agents that will execute modern air-to-air and air-to-surface skills."
In recent years, Eglin has become a hub for work on advanced autonomous systems within the Air Force. The service announced last November that the base had received its first two Valkyrie drones and assigned them to the 40th Flight Test Squadron. It is also set to host six F-16 Viper fighters specially modified to support autonomy testing as part of a project called the Viper Experimentation and Next-Gen Operations Mode (VENOM).
The Air Force has not provided any more granular details about the kinds of tasks the XQ-58A was able to carry out autonomously using this "AI/MIL agent" during the recent test. The service has said repeatedly that, at least for the foreseeable future, there will be a human operator somewhere in the loop when it comes to the employment of highly autonomous drones.
Close coordination between these uncrewed aircraft and crewed platforms is also a key aspect of the service's plans for the CCA effort, specifically. The Air Force is currently planning to acquire at least 1,000 CCAs, if not many more. The 1,000-drone figure is based on a notional concept of operations that envisions two of them being paired with each of 200 sixth-generation NGAD stealth combat jets and 300 F-35A Joint Strike Fighters.
The July fight test does fully reflect a phased approach that the service has crafted to develop, mature, and build trust in AI-driven autonomous capabilities with the expressed intent of getting them out of laboratory environments and into operational platforms.
The algorithms used to enable the XQ-58A to operate autonomously in the July flight test were "matured during millions of hours in high fidelity... simulation events," according to AFRL. They were then validated across 10 flights where they were loaded onboard the X-62A Variable Stability In-flight Simulator Test Aircraft, a heavily modified F-16D Viper, as well as "hardware-in-the-Loop events" involving XQ-58As and other testing on the ground.
The Air Force has explained in the past that this strategy offers immense opportunity to rapidly develop and iterate on the underlying software required for advanced AI-driven autonomous capabilities, and to do so at a relatively low cost.
"We are trying to figure out how to integrate artificially trained neural networks, trained in a simulation... into the real world," Bill "Evil" Gray, the chief test pilot at the Air Force's Test Pilot School, said in a video AFRL released last month, seen below, exploring its autonomous flight testing work. "In this case, to integrate them into controlling an airplane."
"Running these neural networks takes millions and millions... of training runs. You can't do that a real airplane," Gray added. "A flight hour in an F-16 is 10s of 1000s of dollars. But you can do it in a simulator."
"Think of this as the road to future capabilities," Air Force Maj. Ross "WEZ" Elder, a test pilot, said in that same video. "With our highly instrumented systems, we can pull that data out and use it to further mature and develop these systems."
"In the span of 24 hours, we may have trained this thing [the AI agent] many millions of times [in a simulated environment] to do something that we've only seen once or twice in reality," he noted.
In addition, the XQ-58A test last month "builds upon four years of partnership that began with the Skyborg Vanguard and the Autonomous Aircraft Experimentation (AAx) programs," the Air Force's press release added. Skyborg, which was first unveiled publicly in 2019, and the adjacent AAx effort were centered primarily on the development and testing, respectively, of hardware and software to allow various types of drones to operate with high degrees of autonomy.
XQ-58A and UTAP-22 Mako drones, both made by Kratos, as well as the General Atomics Avenger, have been used in the past to support the Skyborg and AAx efforts. The X-62A Variable Stability In-flight Simulator Test Aircraft, a heavily modified F-16D Viper, has also taken part in Skyborg/AAx and other autonomy testing.
More recently, the Air Force acquired at least one MQ-28A Ghost Bat, a stealthy "loyal wingman" type drone developed first by Boeing for the Royal Australian Air Force, specifically to help test autonomous capabilities. The service has previously said MQ-28A and Skyborg are among a group of "technology feeders" it is directly leveraging as part of its CCA program.
It is important to note that the XQ-58A flight test last month is not the first time an AI-driven "computer brain" has been used to fly a real drone under the auspices of the Air Force in recent years. The Skyborg autonomy core system (ACS) has previously flown on a UTAP-22 Mako and a General Atomics Avenger.
In addition, a General Atomics Avenger previously carried out a test flight utilizing an "autonomy engine" developed as part of the Defense Advanced Research Projects Agency's (DARPA) Collaborative Operations in Denied Environment (CODE) program. That technology was subsequently transferred to the U.S. Navy, which is working on its own advanced autonomous aircraft efforts as part of its own Next-Generation Air Dominance (NGAD) initiative.
Though the Air Force and Navy NGAD programs are separate, they are heavily intertwined. The two services are actively collaborating in a number of areas, including technology related to future advanced drones.
Things like Skyborg had also built on previous work on autonomous uncrewed aircraft across the U.S. military dating back decades, as you can read more about in this past War Zone feature.
The July XQ-58A test and the ones mentioned above that came before are also just among the advanced U.S. military aviation autonomy developments that we know about it. Additional relevant work has certainly been and continues to be conducted in the classified realm.
The U.S. military is not alone in its interest in applying AI and machine learning to the aerial combat arena, either. China's People's Liberation Army (PLA), in particular, is actively pursuing the infusion of AI-driven capabilities into its tactical air combat forces.
"AI will be a critical element to future warfighting and the speed at which we’re going to have to understand the operational picture and make decisions," Air Force Brig. Gen. Scott Cain, head of AFRL, said in a statement in the press release regarding the XQ-58A flight last month. "AI, Autonomous Operations, and Human-Machine Teaming continue to evolve at an unprecedented pace and we need the coordinated efforts of our government, academia, and industry partners to keep pace."
Altogether, we can expect to hear about more advanced AI-driven autonomy-related flight testing of XQ-58As, as well as other platforms, as the Air Force pushes to turn these capabilities into realities.
Contact the author: email@example.com