US Air Force wants to pit AI-powered drone against its dogfighting hotshots in battle of the skies next year The US Air Force wants to pit an autonomous aircraft against another fighter jet controlled by a human pilot in July 2021. The American Air Force Research Laboratory (AFRL) has been quietly working towards that goal for a while, and it has more than one ongoing project focused on building unmanned, autonomous aircraft. The …
Fokker Dr.1? Sopwith Camel? A-10? Curious to see what the Air Force brass considers a "fair fight" against a drone, albeit a smarted-up one. Also remember that as the F-35 was being manufactured and put through its initial testing phases, a fly-off between it and the A-10 (the jet F-35 was hoped to replace) was promised. Despite all the huffing and puffing about how great the F-35 would be, when the fly-off actually happened, it was not announced, not hardly covered by the local media, and was reported to have been skewed to the F-35's strengths (-- https://www.airforcetimes.com/news/your-air-force/2018/07/13/a10-vs-f35-close-air-support-fly-off-shrouded-in-secrecy/ --). Maybe we'll get better notice and coverage of the drone -v- dog fight?
A10 v F35, this is a sprinter v decathlete situation, we know who’s likely to win each event before they start. A10s were built for a single purpose, close support, specifically chewing up 1980s Soviet armoured formations in Germany, with the ability to return home having received multiple hit from the AAA systems among those formations, any other task is well outside its comfort zone. The design was heavily influenced by the close support environment during the Vietnam war where it was discovered that fast jets are often too fast to target accurately when the enemy is up close.
Modern weapon systems bring the F35 near to the A10 in terms of close support ability (only ‘near’, not as good), but interestingly, the air force are in the process of re-winging the A10 fleet to provide another 20 years of service (first in service 1977) so it’s clear they’d prefer to keep both for the tasks they’re really are suited to.
The drones that have been employed up to now on military tasks are generally designed for cheap loitering high over a target area against an adversary without any real anti-aircraft capability. Pilot fatigue is real problem in these operations while drones can have their remote controllers literally change shift in flight.
The USAF has been converting actual fighters into remote pilot drones for decades (QF-16 currently) to give pilots dogfight experience ending in a real shot down (the drone doesn't get to shoot), using automatic flight control instead of a remote meatbag here is a fairly easy replacement.
With an AI system acting as the RPV pilot, any experience gained won’t be lost with the airframe just like the Star-Trek Borg, hence the Skyborg tag.
The two biggest problems for the F-35 in current counter insurgency operations are speed and its cannon. The much slower A-10 has a much easier time acquiring targets (especially infantry) and the gun is both significantly more powerful and carries a substantial amount more ammo.
For reference, the GAU-8 fires a 30x173mm round, at 3900 rounds per minute with 1350 round carried aboard the A-10.
The GAU-22 fires a 25x135mm round at 3300 rounds per minute and a mere 180 rounds aboard an F-35.
The fact that the F-35 can't loiter as long didn't help of course, nor does the extremely low availablity of the F-35.
No.
Actually chewing up formations of Vietcong in the jungles of Viet Nam.
The project took a loooong time to get to completion. :-(
But it turns out excellent against the mass tanks of the Warsaw Pact as well. Like many of the sites in Viet Nam they were also protected by radar directed guns and AAM batteries. But the gun and it's milk bottle sized DU shells would do some serious damage to anyone on the ground not inside some seriously fortified bunker.
Doesn't matter.
Drone has serious advantage in dog fight in that you don't have a meat bag that can't take High G turns
So you can design drones to be smaller and more nimble and can't fight human anatomy.
That said... the humans have the ability to think outside the box and may surprise us because the drones will be focused on the norm and not the edge conditions
I think perhaps we're confusing ourselves a bit because of the word "drone". I suspect what is being proposed here is not a rather clunky human controlled surveillance vehicle with a bit of armament attached. Could be. But maybe not. Neither is it a full fledged fighter with HAL or Clippy in the pilot seat.
I suspect what will actually be involved is more like a very smart, reusable, Surface to Air Missile with some high level human control and the capability of returning to base if a target can't be acquired. Such a vehicle doesn't have to win every fight with an F35 or A10. Or even most of them. It's likely to be an order of magnitude cheaper to build and (at least on paper) much easier to deploy and support.
And all the space and weight that you would normally use supporting the meatbag, keeping it alive and comfortable, and giving it a way to control the plane - all of that can be given to some other purpose. Or better yet, cut out of the design altogether.
A drone can pack the same firepower and air-to-air capability into a platform half the size of the F35, and with less than half of the radar signature.
You can do so.
But that might mean that the drone might actually win.
Now is that any way to run a human based airforce?
Unless of course you care more about winning than you do perpetuating your organization.
Which is sort of the point of Pournelle's Law.
Only?
What about being able to handle all the sensor inputs without the need for them to be fused into a single overall sensor picture suitable for human processing?
What about being able to devote 100% concentration to monitoring those sensor inputs, whilst also being able to devote 100% concentration to keeping the aircraft pointing in the right direction, whilst also being able to devote 100% concentration to managing the aircraft systems etc?
What about not needing the aircraft to lug around bulky and heavy life support/other aircrew safety equipment (oxygen generators, ejection seats, survival packs etc.)?
What about not being burdened by human emotions - not hesitating even for a split second before taking the shot, not wondering if it'll survive the fight and return home to the wife and kids?
Being able to pull G-loads up to the structural limit of the airframe is probably the biggest single advantage an AI fighter will have against a human-piloted fighter in a close-in dogfight, but before the fight even gets close enough for G-limits to become a factor the AI fighter will already have been placed at an advantage in the BVR part of the encounter.
The kill factor in WW2 dogfights was the "killer" instinct, the urge to close and make sure of the kill. It's what most of the high-scorers in the RAF possessed.
I suppose an analogy could be programmed into the AI piloting the drone, but I expect a lot of drones would be written off before it was successful. (Though mind you, a drone with no compunction about ramming the opposing plane - and probably with a self-destruct charge to finish the job - would be very useful in Power-to-Power conflict. Not so much in attacking wedding parties.)
It is difficult to make something to regularly withstand much more than 10G - fuel pumps, wiring and stopping the wings falling off. A load of 50G would not be a constant thing - dynamically changing etc.
it likely could be done - just saying that its a completely different airplane. I dont think it would be possible to compare a human controlled jet fighter, and a specifically designed drone . It certainly wouldnt be a done deal for a pilot to beat a drone - today.
How is this going to work though?
An actual close in dogfight? Are those even a thing any more? I appreciate the reason for Top Gun school was to relearn dogfighting when the USAF discovered guided missiles weren't very capable in Vietnam, but a log old amount of time has gone by since then.
If the two aircraft are going to lob beyond visual range missiles at each other its not so much a test of the pilot/AI but one of the missile manufacturer surely?
If both platforms are equipped with the same countermeasures (both being US platforms) that we can assume are effective against the stuff they know about (ie their own) then this too skews the result.
Maybe then the drone is really good at proper dogfighting gun combat?
Which leaves them with what result? The thing is useful assuming the Russian/Chinese radar, missiles and counter measures aren't better/unknown vs US made ones. Might have been true at one point, but suspect that's a big old bit of hubris today.
>>>An actual close in dogfight? Are those even a thing any more?<<
Yes. It's not impossible that modern missile/defence systems cancel each other out to the extent that guns are all that is left, you can't jam or distract a bullet.
Equipment attrition rate is the win / lose factor, when you run out you've lost.
Leaving the pilot on the ground gives a few advantages:
- The drone can pull higher 'G' without squishing or disorienting the pilot
- The drone can be more aerodynamic without the need for a human-sized space up front and a window to look out of
- The drone can be expendable. If the opposition carries four air-air missiles just send up five drones.
Apparently they do, as recently as 2019.
The only problem is that the excerpt also mentions an AMRAAM missile, so it isn't obvious that bullets shot the plane down. Then, of course, there's the politics that muddy the waters further.
But yeah, apparently dogfights are still a thing.
In a real full scale war (WW3) then "kill anything flying" is acceptable with long range missiles without visual verification that it's a hostile aircraft.
Otherwise, it tends to cause problems. For instance MH370 which Russia shot down and has caused them severe diplomatic and public relations problems.
Other near misses go back to the Falklands, where the Argentinian navy had been flying recon missions against the british fleet in Boeing 707's. The Royal Navy had gotten fed up with this, and told the Argentinian's that next time they'd be shot on sight.
Shortly afterwards, an airliner then flew directly over the formation and came close to being shot down, only for the admiral commanding to ask a staffer to get a map and draw it's course and guess the origin and destination. At which point they decided it might be an airliner, and sent up a couple of harriers to have a look and decided that actually, they didn't really want to shoot it down.
Hence, visual confirmation tends to be required before shooting something to bits is authorised, which involves getting close to something that perhaps doesn't want to be shot down, which can result in dogfighting. At least until somebody makes something like a B52 an air superiority fighter with a ton of long range missiles and a couple of tomahawk sized drones for checking out if it's ok to kill something at long range before tossing a Meteor/AMRAAM at it.
Hence the actual military need for a re-launchable surface to air missile.
Launch it at the potential target, have a look at what it is and then decide whether the drone escorts the aircraft then comes home, or not...
However, it likely won't happen because it's not cool and there's no possibility of a meatbag on board. The air force would never accept that it makes them semi-obsolete.
So, what's the drone? Missiles can accelerate and change direction far faster than a plane, good luck in a dogfight with one of those!
If they create a drone for dogfighting that's literally a disposable aerial platform for flying 4+ intelligent missiles to the area, and that has the missiles split up and come at the target from multiple directions then it's unlikely any human pilot would beat it.
This isn't about a computer being a better "pilot" in the same airframe, it's about enabling totally different types of weapon system.
Well I guess if you are just talking about missile carriers it does make sense and probably that is not too hard to achieve.
Cheap, unmanned missile carriers capable of high speeds carrying upwards of 6 BVR missiles, you might achieve better than a 1:1 kill ratio (which is assumed in a mass meeting of opponents both equipped with BVR weapons)
If you look at this chart, you'll see that 3 missiles per target probably ensures a kill:
https://www.ausairpower.net/XIMG/Survivability-vs-Salvo-Size.png
I guess in a protracted war between two tier one players scenario the drone would enable you to win the attrition simply by being quicker to make than a pilot is to train.
And if you are just talking about being able to assemble a wave of them and steer them towards the incoming enemy, I can't actually see why nobody has done this yet?
And if you are just talking about being able to assemble a wave of them and steer them towards the incoming enemy, I can't actually see why nobody has done this yet?
This thinking makes a number of assumptions, which is that everybodies military is roughly equal.
It's not. Basically, western nations control the sky already with fighter aircraft, whereas Russia/China tend to be more interested in mind boggling numbers of tanks etc to steamroller the (land) opposition in an offensive, but have relatively limited air to air capability and make up the difference with prodigious numbers of anti aircraft guns and long range SAM's.
Therefore, our existing forces are already sufficient that the existing arguments even by our adversaries is over if we can have total air supremacy or if we can just have air superiority. Hence, we aren't that bothered about doing it, Russia would probably do it if they could afford to do it, and while China could afford to do it they are mostly interested in keeping us at a distance greater than the combat radius of our aircraft.
Nope. War isn't a tom cruise movie.
When air support no longer exists, if your driving along the ground in a tank/APC then your a more than a little vulnerable.
A single Eurofighter pops up over the horizon, and tosses eighteen Brimstone missiles in their fire and forget mode in a single ripple volley and then legs it. Ten seconds later a tank battalion, it's anti aircraft guns and half a dozen APC's lie burning from a single aircraft, even if your S400 shoots it down when it's on the way back home.
More or less the same thing can be done with attack helicopters, which are designed to barely show themselves over a hill. However, if you have a big airborne radar looking down on the battlefield (even from a distance) then this is a quick form of suicide for the helicopters crews because the helicopter meets a missile from a fighter jet before getting close to it's target.
Simply there is a hierarchy.
Infantry control the ground.
Tanks/artillery wipe out infantry at long ranges. (but are vulnerable to infantry in prepared positions or ambushes)
Helicopters wipe out tanks.
Fighter aircraft wipe out helicopters.
Miss one of these and all the others eat you alive.
S400 out-ranges Brimstone by just how much...I'll clue you in.... somewhere between 5 and 10 times.
S500 is next, even more capable (S400 already effectively being the most capable SAM platform in production). At 600km range they are probably capable of even taking down standoff assets like AWACS or refuelling aircraft.
And of course coming from the heritage their owners come from, its just part of a very complex IADS and not a standalone defence system.
If you were right then you'd be right, but your making the assumption that a ground attack aircraft is going to helpfully wander in at ~30,000 feet which gives a nice clear line of sight to the radar.
If the aircraft comes in at around ground level then the range is line of sight, which due to the curvature of the earth means that even mounted on top of a ten meter pole the radar has an effective range of ~16 miles. Less trees, buildings etc that may obstruct this.
And as noted, the fighter can throw it's missiles and leg it. By the time the missile reaches where the aircraft was at the time of launch then it's not going to be there anymore and if the radar doesn't have a lock then are the missiles going to be intelligent enough to go patrolling for the quickly departing fighter on their onboard radar given that they don't know which way it's gone?
And chances are that there will be some variant of an anti radar missile orbiting to blow the SAM radar to bits as soon as it turns on to make it more difficult for the SAM system.
It's not as one sided for the SAM system as you appear to expect.
I get the feeling that fighter aircraft are regarded as boys' toys and they don't want to have to give them up just because an AI could outfly them, a bit like why cavalry was used in the first world war. Once faced with the reality of cheap, efficient AI driven drones knocking $100 million planes out of the sky such delusions will no longer be an option. In practice air to air fighting is now decided by who can establish missile lock first which will be done way out of the pilot's range of vision.
This article/discussion reminds me of Musk mindlessly spouting off a couple of months ago how fighter pilots were obsolete, and drones were the future etc etc.
It sounds great to the latent science fiction fan in people, but reality as it often does, taps one on the shoulder and asks some uncomfortable questions.
Don't take it from me, I don't have any military flying time (just civil) - but pilots interviewed mentioned quite a few pertinent points - some of which you smart folks have already covered in the discussion (be nice if the article had those, eh?)
Some of these points were taking the advantage in choosing to attack at an advantage before getting to a head to head fight, visual verification of friend or foe, jamming, breaking of encryption and subversion of technology, and lastly human ingenuity - the AI-powered drone is learning as well, and in a fluid fast moving situation, recognizing the better option may still be something the human pilot can do faster today.
Yet there are some important ones in the AI-powered drone's favour: faster time to market compared to training a human pilot, no concern about the human loss, going past human endurance, etc.
I do wonder about the scale of operations though - think the number of alien ships in Independence Day vs. the human pilots.
At some point, all these issues get overtaken by technology - so maybe in a few generations, we'd reach that point. Just not today, or anytime soon.
And yeah, the F-35 is an amazing aircraft in terms of what it can do, but to shoehorn it into every bloody role on the planet, is just hurting its' own cause.
Especially, more so, when the cost of the thing is considered (and the bloody bugs). So, it's not the be all end all standard regardless of what its' manufacturer wants air forces to think.
And yeah, the F-35 is an amazing aircraft in terms of what it can do, but to shoehorn it into every bloody role on the planet, is just hurting its' own cause.
The problem is that if you pick any one of those roles that the F35 can perform then existing specialised aircraft already performs them better. The F22 is better at long range stuff, as is the Eurofighter with a solid state radar upgrade and the new Meteor missile has twice the range. It gets destroyed in dogfighting by literally everything.
ground attack? eurofighter is better from a europeon pov (carries a heck of a lot more) and the A10 is better from a US perspective.
It's a rerun of the situation in the 1970's. If you read Colonel Burton's memiors from the Pentagon Wars (the book is better than the film, although the film is funny given it's a comedy) then there was a fighter program that was supposed to replace everything which ended up so expensive and loaded with hardware for every role that it was crap at all of them, and ended up being replaced by the existing set of aircraft.
The F35 is just a more modern example. The main difference of which is that it's actually made it to the production stage. The crap performance and insane price is about right though.
Most UAVs are designed to hang around at high altitude, observing and targeting ground targets from great distance - they have not been designed for dog fights. Whilst a dog fight could use a modified fighter aircraft, it is worth noting that a considerable amount of effort in designing a fighter goes into keeping the pilot(s) alive. Without such constraints, an autonomous fighter design could ignore human G constraints and lose elector seats, air supply systems, etc. It would be extremely difficult for manned fighter to match the perfomance of a purpose built autonomous aircraft.
Is the win condition the destruction of the opposing aircraft (which would be required to chalk up a win in the war combat situation). If so, when the Natural Intelligence pilot makes it back to the ground, can s|he call it a win? I don't know that I'd volunteer to be that pilot. Youtube video "Take a shufti... don't come back..."
It brings up moral questions too. Already some countries are willing to go beyond what not too many years ago would have been considered immoral, if not illegal acts, provided none of "Our Guys" are hurt or killed. Think of drone strikes to wipe out "terrorists" (and the occasional wedding party) in Pakistan for instance. When there's no consequenses to your act, there is less reason not to do it. It is another step towards the "Willing to kill, but not willing to die" doctrine, but the risk of your people dying is one of the great deterrents to using violence to get your political agenda adopted.
Anyway, as that documentary "Terminator" showed us, the first step to Skynet wiping us out was pilotless F117's with a 100% success rate.
Leaving aside the question of whether or not an autonomous fighter could outperform and out think a meatbag, truly autonomous strike aircraft are more likely to be more likely to strike questionable targets and to create moral distance between targets/victims and the operators.
It's one thing to shoot a man when you can see the whites of his eyes and entirely another to set a mission program for an autonomous aircraft to go and destroy coordinates on a map. It removes the human element even more than drone strikes, which with the current numbers of civilians who seem to have been targeted 'mistakenly' or often with no admission a strike has even taken place, see https://www.thebureauinvestigates.com/projects/shadow-wars .
It also makes plausable deniability even easier; 'Oh it was an autonomous strike , it must have got a glitch...etc', as well as the potential for friendly fire mistakes et al.
Humans fighting humans is bad enough.
Is nobody going to mention this...
Anyway it's only a matter of time until AI can fly rings around humans even if we take physical advantages out of the equation. One assumes they've already got it working in simulators - in fact are there no video games featuring highly skilled AI pilots?
Seems not dissimilar to AI controlled race-car driving - that is proving a hard problem but the gap is closing.
Of course, human pilots are already entirely dependent on computers. Not just basic stuff like constrant trim microadjustments but targeting as well and presumably many other things.
Conventional war in which dog-fights occur is only likely between enemies of more or less equal strength. When there is a huge disparity in military capabilities, warfare is conducted by means of guerilla warfare (aka "5th column", "terrorist"). This type of warfare is thus developed and used by weaker nations (and factions), while the powerful nations try to develop defences against it rather than employing such tactics themselves to any great extent. Increasing your firepower gives you no advantage whatsoever against someone leaving a bomb at a power station or dam wall, or pouring poison into your water supplies etc.
They have a desired outcome - the human pilot winning. The Air Force is full of former flyboys who aren't going to allow pilots to become a thing of the past without being forced to by outside pressure.
That's why it is a 1 to 1 matchup, which is totally unrealistic. The big advantage of drones is that you can make a lot more of them for the same money since you don't have to worry about reliability and pilot survival. Let's see how the human pilot does against 10 drones, or 30. There's no need to spend $100 million on each like the F35, or several times that for the even more expensive (but far more capable) F22.
The real fight these days is before they can even see each other, so a drone (which can be smaller and therefore more stealthy) has an advantage. If it runs out of missiles it can become a missile itself and ram into its target.
"autonomous aircraft" ... so it won't have a remote-control pilot? A la Reapers, and whatnot?
Will it have a Genuine People Personality? Will it write novels and get upset that nobody wants to publish them? Have unlucky love affairs and read bad poetry to hitchhikers strapped into Poetry Appreciation Chairs? Run away from conflict and hide in the deepest basements? (The AI used in the Lord of the Rings movies to animate the Orc armies, gave a lot of them very good survival instincts. They headed away from mass conflicts.)
This obviously needs deep, deep thought ... call the mice and find if they've got any good ideas.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
