No More Pilots
VENOM-AFT
X-62 VISTA
Recently it was announced that an F-16 controlled by artificial intelligence was able to successfully compete with a human pilot in a dogfight. This outstanding feat was achieved after many years of perfecting algorithms to correctly interact with control surfaces, while also remaining situationally aware. The aircraft behind this achievement was the X-62, or as many know it, the F-16 VISTA.
We will take a look at the X-62, its long history, and just what it has achieved.
HISTORY OF VISTA
This would not be the first time the F-16 had served as the platform for research into manoeuvrability. In 1976, one of the two YF-16 prototypes was converted into a test vehicle, known as the Control Configured Vehicle, or CCV. It was designed to research decoupled control surfaces, along with new forms of agility made possible thanks to canards under the air intake. This program would end in 1977, and in 1982 NASA Dryden would begin work on a more ambitious project known as the F-16 Advanced Fighter Technology Integration, or AFTI. This would use an A model F-16, or NF-16A, for its tests.
The AFTI not only continued research into canards and unique manoeuvrability, but in December 1982 it was used to research a voice command system, and by 1983 a helmet mounted sight was being tested. This would later - in 1987 - be integrated with infrared turrets, and a datalink system. It is easy to see how important this program was. Today we see some of these innovations; F-16s are far more capable with datalink and Link16, while the helmet mounted cueing system allows for easier navigation, the placement of points of interest for targeting pods, and the ability to lock targets with high-off bore-sight acquisition.
The AFTI program would continue with research into innovations such as touch screen displays, through till 2001. Much of this research would be picked up by the VISTA program.
The late 80s saw the beginning of the Multi Axis Thrust Vectoring - or MATV - program by General Electric, with designers at General Dynamics, using an F-16 as a testbed. As we have seen in previous videos, thrust vectoring was the research trend at the time. However, the US military initially declined to take part in the project. Thus, the two companies turned to the Israeli Defence Force, who were interested in the program. They were to provide an F-16D for the test, while GE would convert it.
During the same period in 1988, General Dynamics had been contracted to build a unique F-16 known as the Variable stability In-flight Simulator Aircraft - or VISTA. Unlike a standard F-16, this would be a twin seat D model, and in the cockpit would be a variety of changes. Much of the centre panel would be removed, whilst a second joystick would be placed in the middle. New hydraulics were added, along with three new computer systems. This would allow for the flight surfaces to simulate the feel of different aircraft. The aircraft itself was supposed to be a Block 30 F-16D, but it was closer to the Israeli Block 40, with wider fairing on the fuselage.
US Air Force Wright Laboratory bought the aircraft that same year. Later, Wright Laboratory would also approach General Dynamics themselves concerning their MATV program, offering to take part. The Israelis pulled out of the effort in 1991, and Wright Labs took over in a leading role.
By 1992, the VISTA program was in full swing. The aircraft - now designated NF-16D - would take to the sky, performing a variety of agility experiments. Of particular interest was thrust vectoring, which had proven itself to be effective in the F-15MTD. The aircraft would be equipped with a thrust vectoring system known as AVEN. Using an axisymmetrical nozzle, the system had full 360 degree thrust vectoring at a deflection angle of up to 20 degrees. The major advantage of this system is that the thrust vectoring itself occurred at the convergent point of the nozzle and was controlled with a ring which could be retrofitted onto any F-16 using a GE F-100 engine.
The aircraft would be equipped for the MATV program in 1993, and run in that configuration until 1994, with similar research done into high angles of attack using thrust vectoring.
The VISTA program achieved several things during its run. It helped elevate research into thrust vectoring systems which could be fixed to existing engines without major overhauls. The program also directly led to the research and development of several systems now used in the F-35. Namely, the virtual heads-up display, and the direct voice input system which can be used for handsfree application through voice recognition.
Over the next thirty years, the NF-16D would step down from pure research, and be used in a variety of roles, most often as one of the training aircraft at the US Air Force Test Pilots School at Edwards Air Force Base.
X-62 PROGRAM
Now - three decades later - the aircraft has returned to the cutting edge of flight research. In 2021 it was announced that the aircraft would be fitted with new tech, which would replace all the old VISTA components, and adopt the designation X-62.
In 2023 the X-62 made a breakthrough in automated flight; flown by an AI system, it took part in a mock dogfight against another human-controlled F-16. This impressive feat was achieved in a relatively short time as well. At some time in 2022 the AI program was first installed onto the aircraft, and by December the same year it was already flying itself. So, within a calendar year the AI was able to successfully take part in simulated dogfights, reportedly without issue, although two pilots were on board just in case.
Initial flight tests were done from a defensive posture, which the AI handled adequately. Then, after 100,000 lines of code were rewritten, the computer was able to handle an offensive posture with basic fighter manoeuvres, entering into combat at distances of 2000 feet at 1200 miles per hour.
The AI was a culmination of competitions in 2020, in which algorithms competed against each other, and then against an experienced F-16 pilot in a simulator. In these real-world tests, the AI demonstrated a highly aggressive posture. During the simulator phase, the AI understood that safety limits had been placed on both it and the real pilot it was competing against in the simulator, and used these constraints to exploit weaknesses. The reported results show a highly adaptable behaviour. Even with a high closure rate, the AI was able to accurately lead the fight, and would usually take down the other pilot just after merging. With great accuracy, the AI then took down the adversary with a high aspect gun kill.
This makes sense; during and after merging, human pilots must use intuition and look for visual cues to decide where to take the fight. The AI - using an existing database - was likely able to predict turn circle, descent speed, and thrust management in ways not humanly possible, (reminiscent of the AlphaGo system win in 2016 against the world Go champion Lee Sedol). When it came to the kill - which supposedly was occurring just after the first turn from the merge - the AI would make what humans would consider risky decisions, such as immediately bleeding off speed to pull into a high aspect manoeuvre. It is easy to imagine that a human pilot would wait for more cues to be in their favour, and even if they did decide to pull into an extremely taxing manoeuvre, there is a high chance slight human error in the control surfaces would result in an imperfect aim, leading to a slow and vulnerable aircraft. The AI suffers none of these concerns, and lacks no conviction to high-risk, yet accurate manouvers. It was also noted that a separate AI agent also performed well in beyond visual range tests.
This is an impressive feat, partially made possible thanks to the NF-16D, which has once again taken part in a cutting-edge research project. This AI program also highlights just where the future is heading. The cold and calculated algorithm, if well trained, makes few mistakes, knows no fear, and shows no mercy to its adversary. Such AI systems - largely because of their lack of human attributes - may very well replace pilots fastest than we had imagined. Logically, if we dismiss the human desire to pilot aircraft, this is a good option for efficient aerial systems, and the money is certainly flowing in this direction. As ones who don’t want to dismiss the human desire, and attributes, to pilot aircraft, this paradigm shift in military aviation comes with no small degree of angst.
AI offers a unique benefit over other automated systems when it comes to flying. While most autonomous systems require external prompts - such as geolocation using GPS, radar guidance, or other transmitters - a machine learning system can adapt uniquely, even when cut off from such outside information. It can respond to its environment by looking for other external cues, as a human pilot would do, and making decisions based on this information. For example, a wounded automated drone may automatically head back to base and require a particular frequency or cue to land at a particular airbase, on a particular runway, and so forth - code all pre-scripted and non-dynamic to changing situations. In contrast, an AI flying a damaged aircraft can make decisions on the fly, within its own dynamic decision making capability. For example, searching the terrain for what it believes to be a runway, then making an informed decision on how and when to land, just as a real pilot would. It is easy to see the benefits here; A wounded drone in a hostile environment will follow automated sequences, perhaps landing on a runway near to enemy AA fire, whilst the AI controlled aircraft remains situationally aware of dynamic environments and potential threats.
It is not hard to see that there is a new race toward beyond-human-level AI-driven autonomy for military aviation. DARPA’s Air Combat Evolution program, and who knows what other foreign and domestic players and programs, are paving the way toward a future of autonomous air forces.
Will the not-too-distant-future see squadrons without human pilots, and groundcrew led by software engineers?