Autonomous F-16 Fighters Are 'Roughly Even' with Human Pilots | Hacker News Reader

Autonomous F-16 Fighters Are 'Roughly Even' with Human Pilots

nationalinterest.org

47 pointsdapearce47 comments

These aircraft were built with human limitations in mind, take away the human dependency and you can engineer aircraft capable of much higher Gs for longer.

Plus it's probably better to have many smaller and cheaper planes than one big expensive one.

These drones won't be that much cheaper. The life support isn't the most expensive part in high performance fighter jets, it's the airframe, sensors, and engine. The airframe will still have to have some sort of stealth and the engine will be similar to those in other fighters, especially if they're expected to be supersonic too.

Maybe some size reduction? I think they need to figure out just how big enough to carry enough fuel for range and weapons payload

A single sidewinder is almost 200 pounds and GBU-31/32/38 JDAMs are 500-2,000 pounds each, so accounting for the pilot and ejection seat, you're only getting a small bomb's worth of weight.

Making them smaller significantly reduces the volume available for weapons, sensors, and fuel too much IMO. See the Predator, which doesn't have a large weapons loadout and still has the cockpit dome for the radar.

The F-16 was original meant to be cheap, not even having a radar. But by the time it reached production it had far more equipment and weight, for better or worse.

It's original designers thought we should have 10x the aircraft in the sky for the same price. This would result in more fighter pilot deaths, but, as callous as this sounds, maybe that's okay? I mean, it's funny how much expense will go towards saving the life of a fighter pilot while soldiers on the ground can't get proper armor on their vehicles.

Not necessarily, there is a balance to be found. For example, there is no point sending thousands of mini-drones if a simple jammer can make them all fall out of the sky, or if they can't reach their target because their range is too limited, or because they are too slow and visible for a surprise attack.

Modern fighters are big and expensive because of their capabilities like speed, range, stealth, weaponry, counter-measures, radars, etc... And they are not just made to blow up stuff (we have missiles for that), they are also complete platforms for reconnaissance, communication, etc... And the pilot is not just there to control the plane, he is also an officer on the battlefield with all that implies regarding situation awareness and decision making.

That's going to be the ticket for the AI Wingmen. One of the F22's biggest teething problems when it entered service was pilots passing out because the thrust vectoring made it way too easy to pull high Gs in maneuvers. They had to put in software limits and train pilots to be careful not to push it the jets too far.

I often wonder if drones even need to look like they do or if there is a more efficient design that we don't use simply because "things that fly look like this" where "this" includes a place for a cockpit.

I think that’s called a “cruise missile”.

That's already happened, look at how many have a pusher prop configuration to balance out all the electronics and sensor equipment going where the cockpit used to be.

Drones already look different.

The basic design of wings sticking out from a pointy cylinder isn't going to change, because of aerodynamics.

But in the case of the Predator drone, for example, it didn't reduce the volume of the "cockpit" -- it expanded it into a big bulge to fit the satellite antenna and all of the sensors.

We already see in Ukraine that many effective drones are not the "Predator" type that are basically a "fighter jet without the human" design.

Such drones will not last long on the battlefield. The countermeasures for MITM drones is going to be brutal in the next 5 years after what we've seen in Ukraine. Russia has already been doing pretty well using ECM against Ukraine's drones. However, HIMARS has been employed effectively against jamming devices as well.

Low cost autonomous hunter killers as area denial munitions is going to be something else though. Once we no longer require man-in-the-middle these drones will be horrifying. You will be able to just say "Kill anything with these characteristics in this area at this time." Then leave the area and wait for the mayhem to begin. Think land mines that place themselves miles away and then get up out of their hole and then chase down whoever gets near enough.

That plus the cockpit and all that human support equipment. We're essentially looking at high-speed UAVs. I wonder if the speed even still makes sense at that point? Just look at the Ukraine-Russia conflict, once decent enough ant-aircraft systems were setup, it makes little sense for either side to fly them any more.

Instead of a few fighter jets, the future might be a tonne of slow high-altitude UAVs that are always present.

High G is one limitation of human pilots. The added mass and aerodynamic compromises needed for human operation are another. Pilot fatigue is yet another. Training cost is probably the greatest problem. Pilot shortages has to be in the list somewhere as well. This solves a lot of problems.

Yes and no. In principle you could engineer something to support higher Gs but that is not a critical part of the performance envelope in modern combat and it comes at the cost of heavier weight and reduced weapons capacity. Since modern aerial combat is sensor-driven, you'd be better off with a lighter and less maneuverable airframe and larger weapon capacity. You can see this with the recent evolution of combat aircraft being converted into "missile trucks" with massive weapon stores that unavoidably hamper agility.

Bomb trucks is what better sensors/networking enabled. Perfromant AI piloting can enable other tactics, i.e. high performance, high G UAVs with great energetics and endurance can feasibly out evade anti air missiles to bruce force through IAMDs at scale. Sensors would matters less if hardware can't intercept because targetting performance envelop without meatbag limits is harder. Being able to detect =/= being able to effectively engage. Not saying AI pilots have impunity, rather it shifts balance towards offense, i.e. instead of taking 2 missiles to have 95% probability of kill, now need 10 missiles, aka all of a sudden the entire structure/feasibility/economics of all sorts of air defense stops making sense. Imagine mutual air superiority in AI vs AI fighters because it's just too wasteful / impractical to shoot each other down, so everyone goes strait for support nodes on land... or carrier aviation. Imagine a carrier group with DDGs, and air wing designed to deter 100 human pilots now being penetrated/overwhelmed by 20.

Pretty sure you just described Ace Combat 7 UAVs

Higher Gs isn't a significant advantage anymore. Sidewinder missiles hit 60-70Gs and no plane you can make will out pull one. Modern dogfights often last for a single turn or even less. It's often a matter of who sees the other first. Israel's new Python-5 is able to attack jets in a 360 degree sphere and has lock after launch. Countermeasure rejection has gotten so good a merge is essentially a death sentence.

The real reason we are working on autonomous fighters is that we can build jets faster than we can raise and train pilots. It takes 4 years and $5 million dollars to train a pilot, and they will take decades to become seasoned. You can make an F-16 in a few months for $30 million.

High G can significantly reduce no-escape zone of AA missiles with limit fuel for final maneuvering. A performant unmanned fighter with more endurance to play the energetics game can make anti air very expensive, and possibly impracticable depending on context.

High Gs in a BVR engagement aren't that great either. You lose too much energy engaging in that much acceleration. The killer feature of BVR is the trade. If your loyal wingman will happily fly well into minimum abort range and launch high PK shots, against manned targets you can almost guarantee a kill.

>against manned target

Against manned targets sure, US air power can stomp most manned adversaries already. But drive behind autonmous fighters is really about versus against autonomous platforms of peer powers, and really that's PRC, who has AI pilot program of their own. So future scenario that what we'll probably see is a bunch of attritable, high performance loyal wingmans designed to draw as much expensive interceptors as they can against each other, i.e. engagements becomes a unmanned platform attrition / IAMDs magazine depth equation. Question is, who wins in that numbers game? I would say land aviation with better logistics vs carrier aviation with limited magazine depth.

Then what does that mean when this tech proliferates, especially local vs outside power projection - because autonmous piloting removes expensive intitution building layer and will enable many shit-mid tier powers to now have airpower. Maybe even very competent air power depending on training data.

Personally I think it will be more like a bunch of slow, high altitude, cheap drones with high performance BVR missiles and datalink. The missiles have just gotten too dang good in the last 20 years. When the MAR is over 50 miles on the deck, energy fighting just doesn't matter anymore.

This is absolutely correct. The air-to-air game is very complex and high-G is still a thing.

And yet, this is small ball stuff. When I think of all the possible ways autonomous combat aircraft will revolutionize air combat it's a bit overwhelming. We're talking about unlimited, fatigue-less, highly effective, fearless piloting of aircraft that can assume designs that don't have to accommodate or risk a human.

If it isn't yet an arms race, it will be when some conflict sees someone punch way above their weight because they have the supersonic robots and the other guy doesn't.

>they have the supersonic robots and the other guy doesn't

Should also consider what happens when more actors have supersonic robots if/when this tech proliferates. AI pilots is going to lower barrier for _real_ airpower for small/medium size powers. Not just US is training relevant models. And AI pilots = accepting future is going to be full of attritable platforms, whose going to be making/selling all the cheap disposable defense hardware. Just like drones, commoditizing theatre level air power is going to be gamechanging.

Pilots can survive maneuvers that damage the airframes, it isn't just a free for all.

I have long maintained that command and control aircraft will be the only piloted things in the future (and just so they can evade easier). Have them stand off and be able to control directly via line of sight/hard to jam comms (likely by relay of some comms drone network), but also have relays back to other controllers in case the primary command and control becomes unable. By 'control' though it is clear that this will just be authority based and much higher level 'patrol over here weapons free' not 'turn left to 180'. The future is drones, either single use (missiles, TLAMs but with more autonomy, etc) or multi-use like reapers, etc but all of them with much more autonomy.

The fighter pilot mafia mentality is hard to change though just like the bomber mafia. Ukraine is likely changing that mentality a lot right now though. This feels like a battle ship vs aircraft carrier moment for aircraft so maybe this shift will happen sooner than I thought. Not that that is a good thing unfortunately, just an inevitable one.

This is quite a dangerous mindset, pushing for AI weapons with limited constraints:

> “I think the future is becoming clearer,” Kendall continued. “I think the only question that really may remain is who’s going to get there first? And what are the constraints we want to place on ourselves that will limit our operational effectiveness compared to our adversaries and how we manage our way through that.”

It has some resemblance to the arms race to develop a nuclear weapon. Several people were exposed to radiation, tests killed civilians, all because both sides were afraid of not being the first.

I wanted to make a joke how Sopwith [0] enemy fighters were already good in the 80s but this is kinda terrifying.

Why are they trying to retrofit an F16, a whole new platform without human-required accommodations would make more sense..

[0]: https://en.wikipedia.org/wiki/Sopwith_(video_game)

Do you know how much time passes between "let's make a whole new platform without human-required accommodations" to an even an aerodynamic model? Not even talking about the first flight.

> Why are they trying to retrofit an F16

Because there are lots of them,

As of 2023, it is the world's most common fixed-wing aircraft in military service, with 2145 F-16s operational. [1]

and several countries are phasing them out in favor of the F35. Pilot training is a slow and expensive affair (ask Ukraine), so it makes sense to concentrate scarce resources on the top model - unless the legacy one can fly itself.

[1] https://en.wikipedia.org/wiki/General_Dynamics_F-16_Fighting...

If you have a bunch of cars sitting in your garage and you want to test your auto-driving software, you don't start by building a car. Simple as that.

Seeing term "loyal wingman" - meaning an unmanned combat air vehicle - makes me ask: Are the human pilots not considered loyal enough? Maybe so. The humans might hesitate on an order to shoot whereas the AI will not. I prefer the human that at least hesitates to shoot at a target that will cause substantial "collateral damage".

I think it's in reference to delegation 'wingman' and availability 'loyal'. I like the Russian word Sputnik for 'fellow traveller' but obviously that would not be appropriate for US weapons.

Missiles, which are essentially single use "unmanned combat air vehicles" don't hesitate to destroy whatever target they are shot at.

If these "loyal wingmen" only shoot when ordered to by a human, these can be seen as reusable missiles more than actual human wingmen.

loitering recoverable munitions.

There is a spectrum of expected reusability, even with human piloted aircraft.

Historically at least, one of the significant jobs of a wingman has been to act as bait. With dogfighting, with cannon fire, fighters are most vulnerable when they're trying to shoot down another plane, because they need to be pointing to their target so they tend to be going in a fairly straight and predictable direction. The idea was the wingman would lure one of the enemies onto their back, and then the their partner would dispatch the relatively easy target. This is one of reasons why fighter aces in WWII had such a lopsided kill count compared to everyone else. In order for this all to work, the pilots have to trust each other, and it was certainly the case that some fighter aces put a higher priority on getting kills than preventing the death of their wingman.

I'm not terribly knowledgeable about these things (and most of it is theoretical anyway because most of this tech is untested), but I suspect that there will be a similar mechanic with modern stealth and beyond-visual-range air warfare, whereby firing missles makes you easier to locate and ultimately target, so having a wingman who's willing to act as target practice while another pilot waits for an enemy to betray their location could be useful.

related, about trust in this context: Is the Pull-Down Effect Overstated? An Examination of Trust Propagation Among Fighter Pilots in a High-Fidelity Simulation https://journals.sagepub.com/doi/10.1177/15553434231225909

I think what is important to note is that fighter jets already make use of highly sophisticated and automated weapons. E.g. Air-to-air missiles, which can accurately track and hit a target over extremely long distances.

It seems quite unsurprising that humans don't excel at flying modern fighter jets.

The Air Force’s Collaborative Combat Aircraft program is developing autonomous aircraft that are likely to be part of the larger NGAD effort.

It would consist of the unmanned, loyal wingmen that would support a manned fighter.

How long before that's flipped and the manned fighter is supporting the UAV?

I think the idea is that the pilot in the manned vehicle is supposed to be the one to decide on weapons employment. Also it's much harder to jam communications at sub-mile distances between a pilot an their LWMs.

If you flipped it, you may as well omit the manned fighter.

Faster ODA loop, much shallower incorporation of energy tactics and greater strategy. F-16 is not an energy fighter though, it's a dog fighter (at least in original design). Not surprising to see much faster ODA loop win out.

Do you mean OODA loop? Observe Orient Decide Act?

Oof. Yes. Thanks for the correction.

Finally some really cool news.

Especially since this is a dogfight and not an air-to-air missile exchange.

Would be interesting to know if the AI came up with new maneuvers.

Reply on news.ycombinator.com