General Atomics Aeronautical Systems, Inc. (GA-ASI), has revealed that its jet-powered MQ-20 Avenger drone, equipped with U.S. government-provided autonomy software, took part in a recent Orange Flag large force test exercise. This is the latest milestone for the Avenger, which has been heavily involved in the testing of different artificial-intelligence-driven autonomy systems for several years now.
GA-ASI announced today that one of two company-owned MQ-20s, with the so-called ‘reference autonomy stack’ software, flew in a demonstration at Orange Flag 25-1, which took place at Edwards Air Force Base, California, from February 19-21. You can read more about the Orange Flag series, which is focused on developmental test activities, in this previous article.
A General Atomics MQ-20 Avenger unmanned vehicle returns to El Mirage Airfield, Calif. June 24, 2021. The MQ-20 successfully participated in Edwards Air Force Base’s Orange Flag 21-2 to test the Skyborg Autonomy Core System. (Photo courtesy of General Atomics)(This photo was edited by blurring the tail number for security purposes.)
A General Atomics MQ-20 Avenger unmanned vehicle returns to El Mirage Airfield, California, on June 24, 2021. The MQ-20 was participating in Edwards Air Force Base’s Orange Flag 21-2. Photo courtesy of General Atomics Air Force Test Center
Once integrated in the MQ-20, Shield AI’s reference autonomy stack — which included a pilot vehicle interface (PVI) — demonstrated autonomous flight operation, with a focus on air-to-air engagements.
“The government-provided PVI enabled seamless control and monitoring of the autonomy stack, highlighting the interoperability and flexibility of GA-ASI’s UCAV ecosystem,” the company said in a statement. “The Shield AI stack demonstrated autonomy skills for safe administrative phases of flight.”
“This demonstration marks a significant achievement in our ongoing efforts to operationalize autonomy for UCAVs,” added GA-ASI Vice President of Advanced Programs Michael Atwood. “Flying the government reference autonomy stack at Orange Flag 25-1 and utilizing the government-provided PVI underscores our commitment to delivering robust and adaptable autonomy solutions for the warfighter.”
An important part of the demonstration was to prove that the MQ-20 — and, by extension, other GA-ASI drones — can be rapidly reconfigured from using company-written software to government-provided or other vendors’ software as required. This has important implications in terms of flexibility for future drones once in service, so they can receive new software — enabling new capabilities and/or enhanced interoperability — very rapidly. Speaking to TWZ at the Air & Space Forces Association’s 2025 Warfare Symposium, a spokesperson from Shield AI said that the process of integrating the reference autonomy stack in the MQ-20 took “about three weeks.”
A picture of a drone control system using a tablet-like device that General Atomics has previously released. GA-ASI
The demonstration also showed that the MQ-20 can rapidly swap between autonomy systems mid-flight, as C. Mark Brinkley, a General Atomics spokesman, explained to TWZ. “During Orange Flag, we were able to fly both a government [AI] architecture as well as sort of mid-flight switch over to the Shield AI software and let that software take over.”
Brinkley also told TWZ why he considers the MQ-20 an ideal platform for these kinds of tests, which are intended to feed into the Air Force’s Collaborative Combat Aircraft (CCA) program.
“The interesting thing about MQ-20 is it’s sort of still ahead of its time as far as jet UCAVs goes. We continue to use it as a CCA surrogate, and I suspect we will continue to do that for some time. We have two company aircraft that we own and can fly and use whenever we want. The ability to use the MQ-20 the way that you would a CCA in terms of loading on software, experimenting with autonomy actions, experimenting with different types of missions, like we did in Orange Flag, and then be able to take all that data off of a real airplane, and go back into the virtual world and make changes as necessary, it’s really been a game-changer for our ability to move that forward.”
“For us, it’s really using an aircraft that exists,” Brinkley added. “We don’t have to wait until there are some new aircraft. I mean, we’re building our CCA. We have plenty of aircraft out there. Having two company-owned MQ-20s, you might look at them and say, ‘Well, that’s too big for what the CCA program might be, they might cost too much. It’s not exactly what you’re looking for.’ That’s fine. But in terms of trying to get to autonomy and autonomous flight and autonomous missions, you can do a lot of it in the virtual environment, but really having a real aircraft to team with them has been unique for us.”
In the context of the MQ-20 demonstration at Orange Flag, the drone’s reference autonomy stack was a product of Shield AI, a relatively new by fast-moving aerospace company that you can read more about here. The stack is referred to by that company as Hivemind Enterprise.
Shield AI’s original Hivemind was an ‘AI pilot’ that has been used in real-world fighter aircraft test flights linked to advanced U.S. Air Force programs, as well as to control a host of Kratos drones and in Shield AI’s own MQ-35 V-BAT drone. Hivemind is also expected to feed into the Air Force’s emerging Collaborative Combat Aircraft (CCA) program.
A V-BAT is prepared for flight onboard the destroyer USS Michael Monsoor (DDG-1001) during Rim of the Pacific (RIMPAC) 2022. U.S. Navy
In the words of Shield AI’s co-founder, president, and chief growth officer Brandon Tseng, Hivemind “enables aircraft to accomplish missions fully autonomously without GPS, without communications, without a remote pilot. What do I mean by fully autonomous mission execution? In the example of our NOVA quadcopter, it would go inside buildings, and it would find threats, it would map out the building, it would clear the building all by itself without anybody piloting it. GPS didn’t matter, communications didn’t matter, it would accomplish the mission.”
The NOVA 2 quadcopter, which features Hivemind. Shield AI
From the start, Hivemind was intended as a core central system within an aircraft and one that can be leveraged across many different platforms. In this way, Hivemind first began to be installed in small quadcopters and then in increasingly more complex drones, all the way up to the MQ-20, as well as crewed surrogate aircraft.
In addition to providing ‘self-piloting’ technology for aircraft, Hivemind enables cooperative teaming and swarming. Using Hivemind AI, a group of drones can execute a mission, working together dynamically, reading and reacting to each other, to the battlefield, to the adversarial threats, and so on. This can be accomplished at computer speeds.
Hivemind is designed to be agnostic to the original equipment manufacturer, and the company has said in the past that it wants the AI technology “to fly on every single CCA.”
However, Hivemind is just one of a number of AI agents that has been flown in the MQ-20.
“We’ve proven that we’ve been able to fly a lot of different autonomy engines and autonomy pilots in the aircraft,” Brinkley adds. “It’s flown Skyborg, it’s flown the DARPA CODE engine. It’s flown our own AI.”
Brinkley is also cognizant of the fact that AI software from more than one supplier might find its way into the CCA program and into other GA-ASI drones.
“With software, it’s not a winner-take-all kind of game. Just like when you take your phone out and you’re running an operating system on your phone, but you and I might have different maps, we might have different email programs, we might have different games that we play or whatever. There’s not going to be just one way to do this. Across the services, across different nations, they may all want to do things differently.”
Whoever provides it, AI will be especially important to achieving the Air Force’s ambition for crewed-uncrewed teaming, including having fighters, like the F-35, operate as inflight ‘quarterbacks’ for these advanced drones. For the Air Force, how pilots in crewed aircraft will actually manage CCAs during operations has emerged as an increasingly important question and has already yielded related trials, as you can read about here.
The Air Force is also still very much in the process of developing new concepts of operations and tactics, techniques, and procedures for employing CCA drones operationally. How the drones will fit into the service’s force structure and be utilized in routine training and other day-to-day peacetime activities, along with what the maintenance and logistical demands will be, also remains to be seen. Questions about in-flight command and control have emerged as particularly important ones to answer in the near term.
Models of the drones General Atomics (at top) and Anduril (at bottom) are developing as part of the CCA program’s Increment 1. General Atomics/Jamie Hunter
The latest edition of the Orange Flag test exercise further underlines the importance of the MQ-20 as a testbed for future autonomous collaborative platforms and adds yet another platform to the growing roster that is using different iterations of Shield AI’s Hivemind AI technology.
Contact the author: thomas@thewarzone.com