This is an incredible video. Hearing the pilot gasp for air while a) his wingmates yell on the radio and b) seeing the details on the HUD are amazing.
The airspeed goes above Mach 1 while the altitude drops from about 16,000 ft to 4,000 ft. The pilot sustains over 9G when the automated system pulls him out of the dive.
This looks like it was an air combat exercise. At the very start of the video (0:07) you see a big round circle and some lines that are spread a the top and close together at the bottom. That's the EEGS gunsight, and in the censored portions of the screen is likely the bandit/hostile he was tracking. (Exercise of course.)
Either his target slips out of gun range or he switches to a different mode, but regardless it looks like he pursues, rolls 90 degrees and
"buries the stick in his lap" (F-16 has a side stick that doesn't move, but whatever. =). I'm told you have to be careful with this, and probably in the F-16 doubly so. (As C.W. Lemonine has said in a few of his videos, the F-16 will try to kill you.) In the top left of the HUD, straight up from the "C" is a decimal number showing the Gs on the aircraft, and it climbs rapidly from 3-4G up to 9G. He loses about 100knots of airspeed (still pretty fast though!) but then you hear that exhale around 0:18 and he starts to slip off the horizontal and enter that dive. His 9G turn probably is what knocked him out, and yeah, breaks Mach 1 briefly as he heads downward.
Then as ya said, he pulls another 9G when the system pulls him out of the dive.
The "knock it off" calls after are to indicate that they're not fighting anymore. As I understand, that's not slang, that's the actual terminology used.
> The "knock it off" calls after are to indicate that they're not fighting anymore. As I understand, that's not slang, that's the actual terminology used.
Yeah, AIUI that is essentially the safe word (phrase) that anyone can call at any point that says "everyone involved stop now".
You hear SULLY1 call "SULLY knock it off" (command) and then call out his own confirmation/the now conscious pilot calls the confirmation as well.
From what I've read in the past, Auto GCAS computes a 5G recovery pull. Speculation was that the pilot came to consciousness at some point during the recovery and applied some extra input to hit 9Gs.
Pilot recovered while pulling 5G? I'd think they would have designed the GCAS to avoid a crash at all costs, regardless of the Gs required. Crashing is worse
It is designed that way. It realizes that by defining a roll to level and +5G climb as a successful recovery, and preventing the aircraft from being flown into a condition where a successful recovery is impossible.
So if you're flying right above terrain and you suddenly fly down, the airplane should accept your inputs and fly towards terrain until the moment recovery is impossible and will then engage.
Agreed, even if he was conscious during that nose dive, pulling 9.1 G’s would send him right back into GLOC before fully recovering.. and likely into the ground as he was already so low...
Only this computer system could’ve pulled off such a save. Awesome!
how can you read that? I look at your comment and look at the HUD and can't match up any of the numbers and axis to your observations? how do you read this?
They probably used to play flight simulators (I did). The vertical ticker on the left is airspeed in knots. Ticker on the right is altitude in feet. Then underneath the left ticker find the word SIM, directly beneath it is the plane's current mach number, and diagonally below and to the left of that is a counter that tracks the highest G load the plane has sustained (it goes to 8.4 right as the pilot passes out then peaks at 9.1 during the pull out).
The repeated "Recover!" is indeed the formation leader, not an automated warning. It's as regular as it is because pilots, and especially military pilots, are trained to maintain a very strong radio discipline, and also (as Brian Shul notes in Sled Driver, his SR-71 memoir) because as a point of personal and cultural pride everyone wants to sound like Chuck Yeager on the radio. It makes sense, too; things can go wrong very fast in an airplane, and it does no one any good for a pilot to lose their cool in an emergency.
Listen closely, and you'll notice that the formation leader lets that discipline fray a bit in the second and following calls, when some urgency bleeds through. This is because he suspects his trainee is about to die, which is what would indeed have happened if AGCAS hadn't been available.
(You can also tell it's a radio call and not an annunciator by the form of address - the full call is "Two, recover!"; "Two" is what someone else in the formation calls its second member. "One" is the formation leader; in a four-ship formation you'd also have "Three" and "Four", and so on for larger groups.)
Yep. The actual AGCAS annunciation is the pair of chevrons advancing from the sides of the HUD into the center, and the blinking cross and "FLYUP" indicates that the system has engaged.
For audio, there's a very noticeable horn during recovery, as well as a "Fly up" automatic voice announcement - that's a little hard to pick up due to overlap with a call from the formation leader, but you can catch it if you listen closely.
Question: What repercussions does a pilot face after an incident like this? Is there more physical training done do the breathing maneuvers, and plane training to prevent excess G’s like this? Is the pilot grounded for a certain period of time?
wow, this is making me tear up for some reason. thanks for sharing. i guess i've never seen such a clear case of tech making a life saving impact before.
Software is the implementation, but this is a large an nuanced program. I find the most interesting aspect is that this is software specifically ignores the pilot. If the g-locked/unconscious pilot is slumped over the controls the computer will ignore his stick inputs and substitute its own best judgment of the situation. (Pilot's stick says down, computer decides to ignore and pull up instead.) That is a really powerful change in pilot culture. It conflicts with the 737-Max fiasco where we all screamed that pilots should have final authority over computers. Getting this into f-16 cockpits must have involved many very heated discussions.
Not as much as you'd think. During early evaluations, AGCAS proved itself a lifesaver. I'm pretty sure there is a way to override, if it malfunctioned. There are cases where the system falsely detects an unsafe condition, the pilot hears the warning tone, before AGCAS takes over.
One thing the 737 Max doesn't have, which the F-16 has, is ejection seats. The pilot can eject if AGCAS puts them in an unsafe situation. I don't believe there has been a single case of a pilot ejecting due to a false AGCAS activation.
I'm not an expert, but I think F-16s were designed as fly-by-wire from the start, while 737s had mechanically-linked controls. This probably made it an easier sell on the F-16 case.
Indeed. And another thing to consider is that some modern 4th gen performance aircraft require that fly-by-wire by design, as they're designed with inherent aerodynamic instability. You'd rapidly lose control of the plane if there was a FLCS (FLight Control System) failure. Checking status of these is part of the startup sequence.
These systems are often triple or quadruple redundant, and will prevent the pilot from executing some maneuvers if it would put too much load on the aircraft. This can be relaxed for say, wartime situations, as placing high-Gs on the air frame reduces it's lifespan.
The article makes it sound like all Block 40 and newer F-16 have AGCAS. While this might be correct, it's worth mentioning that AGCAS was implemented around 2014 and started rolling out on F-16 after. Block 40/50 F-16s are way way way older than 2014 ... plus most non-US F-16 units didn't and won't get the update at all.
Similar story for the F/A-18, but even later. The technology exists—Boeing even did a demonstration in the early/mid 2010’s. Navy only finally approved installing it in FY2019 budget, so it may not even be fielded yet.
I used to work with a Marine F/A-18 pilot who came to the program office with an axe to grind about this. He requested meetings with PMA-265 (not our office, but same building—this was a very weird thing to do) to _politely_ show them the USAF F-16 AGCAS video and a list of folks he personally knew who died because we didn’t have AGCAS in F/A-18. I just looked him up, and he’s been promoted to Lt.CoL.; now working as the air systems lead for legacy F/A-18 in PMA-265–and it sounds like even legacy hornet is going to get AGCAS. I’m impressed (but perhaps not surprised) that he’s been able to get them to invest AGCAS in a platform that hits end-of-life in 2030. 11/10 would work with SOCK again.
Huh, nice to see something that I imagined would be introduced eventually, but didn't expect to arrive so soon.
A while ago I came up with an idea for a short sci-fi story and it goes something like this:
An AI assisted fighter jet is being developed which not only recovers from situations like in the article, but also performs some of the fight-related duties - often faster and better than a human pilot would - learning as it goes.
The top brass is impressed, so the fighters are eventually sent on their first mission.
Enemies are engaged and dispatched one by one, but two things seem off: the planes start making increasingly sharp(and effective) turns and contact with the crew becomes limited at first and stops altogether later on.
As the last enemies start falling back the unit begins pursuit - even though they were ordered to return to base.
No response. Meanwhile the aircrafts' movements become even more erratic - to the point where eventually they fall apart one by one from the sheer strees put on the parts.
Recovered blackbox recordings indicate that by the time the enemies started retreating the pilots were all either unconscious or long dead.
This reminds me of an episode of Ghost in the shell (the serie, season 2 episode 4).
from the description of the episode:
the pilot of the helicopter suffers a massive heart attack, and the Jigabachi begins to spin out of control. The on-board Artificial Intelligence gets the helicopter back under control, and the military officers running the drill decide to abort it out of concern for the safety of their troops. However, the AI aboard the Jigabachi refuses to acknowledge the order to return to its parent JMSDF aircraft carrier. Acting on the assumption that the chopper is under attack, the AI overrides the flight controls of other armed Jigabachi aircraft, an air tanker from the carrier, and several nearby military bases, ordering these units into a tight defence formation in the heart of the Niihama Refugee Residential District.
Heartily disagree. Half of the serie first season (the complex episodes) is somewhat interesting. The rest is mostly the usual sci-fi trops and goes from mostly avoidable (the other half of the first season) to plain bad (everything after). It has neither the depth nor the artistic chops of the two Oshii movies.
I agree. The tv series is better because with 50+ episodes, the creator(s) were able to present a more detailed version of the future described in the movies (I'm talking about the first one from 1995 and Innocence).
At this point the WWII dogfight between planes is dead. Missiles have more than enough power an maneuverability to take down any plane their are pointed at. There are ways to deal with missiles, but none are maneuvers by the target. You can maybe out fly a missile, but that is because it was launched at close to maximum range and so you get out of range.
Drones are a different story. The future is pilots on a different continent from the actual fight. They will sometimes take the controls, sometimes push the "do this maneuver" button, and sometimes let the AI take care of it. As AI get better and better their role becomes more as the final human in the loop agreeing to kill an identified target. There are still gaps in making this work in the real world, but the signs are all there. Without a human in the plane you can do things that would kill the human.
Your first paragraph isn't accurate. For example, one common maneuvering tactic against radar guided missiles is to fly perpendicular to the radar, trying to hide in the doppler notch.
Against IR guided missiles, maneuvering is also a part of the counters used (in addition to flares). Maneuvering helps deplete the missile of energy. You'll never out turn a missile, but you can make it run out of fuel/energy.
That said, modern AAMs have tremendous no escape zones where it's quite difficult to survive if the missile is employed properly.
And drones flying counter air missions will probably happen when level 5 self-driving is successful.
> There are ways to deal with missiles, but none are maneuvers by the target.
Countermeasures combined with maneuvers to break lock and prevent reacquisition is, as far as I know, still doctrine because it has a high enough probability of success to be useful (and to be worth equipping planes for).
I wonder how many WW2 (the eastern front kind) and early-jet dogfights were won not because of airplane damage, but because one of the pilots GLOC'd himself and hit the ground.
Or just exhaustion. There's an interview with a Japanese ace Honda Minoru on yt in which he talks about how pilots were so exhausted from flying eight hour combat missions every day of the Guadalcanal campaign that they would fall asleep at the controls, slowly fall out of formation, and hit the water. They (amazingly) had no radios in their aircraft at the time, so they just watched it happen.
IIRC they did have radios but those were so unreliable due to poor quality of vacuum tubes that many pilots chose not to use them, sometimes to the point of removing the antenna mast.
The Japanese had a variety of issues with radios. Some Type 0 pilots did remove their radio system, to save weight, but only on land-based fighters. For carrier-based ships radios were still essential, because they used radio direction finding for navigation. Japanese carrier-based operations also suffered issues because they tended to use a single radio frequency for all air operations (meaning the channel could get disorganized). Japanese carriers also had their antennas on the sides of the ships rather than at the top of the island, which meant they often couldn't receive longer ranged transmissions, so it would be up to escorting cruisers to receive messages and then transmit them to the carrier. The book "Shattered Sword: The Untold Story of the Battle of Midway" has some good information about some of these nitty-gritty details.
I'm pretty sure I've read that exhaustion was a factor in the Battle Of Britain - the RAF took measures to ensure that pilots were reasonably well rested between sorties whereas the Luftwaffe just kept throwing them in again and again.
The Germans tended to put their trainers into combat roles meaning that they had trouble training new pilots. In the later days of the war they sometimes told pilots to eject if they saw an enemy plane over friendly ground - there were plenty of airplanes so better to make it back alive and try again than to risk death.
The Japanese did the same thing, toward the end of the Pacific War. It wasn't by preference in either case, but because they were so low on pilots that it became a question of whether to use instructors as combat pilots or just not have anyone to fly those missions at all.
Dive bombers actually subjected pilots to some serious g-forces. Some planes (at least the Stuka) had an automatic pull-out that would cause it to pull up at a certain altitude (or when the bomb is released, I'm not sure), so the plane could recover even if the pilot blacked out. Though apparently some pilots didn't like this feature and disabled it, because they felt always recovering at the same height made it easier to anti-air defenses to target them.
Another fun fact about the Stuka: you know that weird whining noise you hear in WW2 movies when a plane is dive bombing, almost sounds like the engine is acting up (like this https://youtu.be/5uvqhA4_2tU?t=39 )? So that noise is unique to the Stuka. It's not the engine, the plane has sirens fitted to the dive brakes! It was meant to scare soldiers on the ground.
I've definitely heard stories from old pilots of blacking out and coming too while out of control, and then having to wrestle control back while coming round. I wouldn't be at all surprised if many more never recovered in time.
I remember a WWI era flight-sim called Red Baron that included this in it's mechanics. If you tried to pull a turn too fast you'd start to black out and lose control.
From this talk [0] I get the impression that a large part of the "chess game" of a dogfight is about the aircraft's momentum. You're trying to get the enemy into a situation where his aircraft has too much or too little energy to respond to you, and can't accelerate / decelerate in time.
As early as it was, their construction was probably sturdy. It’s only now that we calculate material margin at 110% instead of x5 or x10. The B52 is so stretched it famously leaks fuel at ground level, the operating guideline says “6 droplets per minute” to 20 droplets per minute for a dozen points of the aircraft (mostly around the wings).
Why only the eastern front? Pilots of the RAF were completely exhausted during the Battle of Britain and the Blitz since they had such a numerical disadvantage compared to the german air force.
Especially when you consider that most of the German Luftwaffe were jacked up on "pilot's salt" (meth) [0]. Makes sense that the US AirForce now has a widely known Modafinil kick ;)
Not just the Luftwaffe, either, and not just speed. Tablets formulated with both methamphetamine ("Pervitin") and, later, cocaine, were issued throughout the Wehrmacht. [1] Some branches also had speed-laced chocolate bars, Fliegerschokolade and Panzerschokolade for pilots and tankers respectively. [2] They also broadly issued opioids - including oxycodone ("Eukodol")! - and drunkenness was likewise extremely widespread, at varying times both with and without official sanction and material support (i.e. liquor rations).
Thanks for your reply, I had no idea the germans issued go-fast chocolate bars! The implications of alcohol rations are also interesting - especially for the social bonding aspect the germans sought to exploit among their troops.
I don't know so much about that, especially in that the Wehrmacht inherited the Prussian tradition and thus generally looked with severe disfavor on vertical fraternization among the ranks. The Waffen-SS maybe, but they also always struggled to be taken seriously by the Heer, who not without cause regarded them by default as dangerously incompetent pretenders to the profession of arms - some units and formations eventually
became very effective in combat, but many didn't, and Heer commanders frequently complained about and sought to avoid operating with the Waffen-SS overall. The Heer's contempt also had much to do with the conflict between the aforementioned Prussian military-aristocratic tradition of their officer corps vs. the SS's origins as an offshoot of the "working-class rabble" of the SA, with its reputation for fraternization-driven collapse of discipline so severe that the whole organization had had to be purged. (We know now that this isn't really true; it was the SA's broad populism and redistributive political tendency, and the concern these caused among the corporate elite Hitler and the Party establishment had so carefully cultivated during their rise to power, which drove the purges - debauchery and indiscipline were mainly just a convenient propaganda cover, but the propaganda worked and these were understood at the time to be the driving cause behind the purge.) So, for political reasons alone, it seems unlikely the Waffen-SS had much use for liquor as a bonding tool. Too, the SS overall had their own weird "New Aryan Man" ideology thing going already, which tended to frown somewhat on alcohol even before that was reinforced by party leadership about halfway through the war. And finally, as social bonding promoters go, I suspect booze doesn't add up to much next to the total experience of shared purpose and adversity that is WWII-style mobile warfare.
I can think of a few reasons why liquor might feature heavily in the WWII German experience of war. Primary among them, I think, has to be that it acts as something of an emotional anesthetic, and soldiers in both wars often used it to help cope with what would otherwise be intolerable. In WWI that was mostly the manifold horrors of life in the static trench warfare of the Western Front - you see something similar in WWII Stalingrad, for example, though only sporadically and briefly due to the ever-straitening circumstances of the besieged Sixth Army. Put simply, they ran out of everything before Paulus finally threw in the towel, but while they still had liquor, this was the way they used a lot of it.
Alcohol also helped support the rapid, continuous advances required in the WWII style of mobile warfare, serving as something of a dual to benzedrine. The ferocious German materiel buildup of the 1930s notwithstanding, their entire war plan, again much as in WWI, was predicated on the knowledge that their only path to victory lay in finding a way for a smaller force to beat a much larger one. Technical and training superiority was one aspect of the solution; another was the speed, precision, and decisiveness of action that that individual superiority enabled. All of those grow steadily harder to maintain over time, as action takes its toll, and countenancing alcohol use helps blunt this effect for a while. In the long run I'd expect it to be more a hindrance than a help, but German plans in both wars were intended to ensure that the war was won before there could be a long run - because, in the long run, the Germans knew they would lose.
And, of course, alcohol helped blunt the psychic damage of participation in atrocity, for the vast majority of soldiers and others to whom it did not come naturally - this, along with suicide, was in particular a problem among early Sonderkommandos and prior to the industrialization of massacre for which the Nazi regime is most deservedly loathed today. Part of the purpose behind that industrialization was in fact to provide enough emotional distance, for those tasked with carrying it out, to stop them constantly drinking themselves insensate or eating their guns or both.
I would guess that the presence of AGCAS is, in part, why the pilots were willing/allowed to push themselves so hard in training.
I didn’t really say if the system only intervenes if it believes the pilot is unconscious or if it always intervenes if the fligh path is within some envelope of terraforming.
I doubt they changed their training methods due to AGCAS.
From my understanding, system works automatically and does not take into account whether pilot is conscious or not. It is not meant just to save unconscious pilots but also those who are disoriented (in fog, confused, ...).
Given that it is a military aircraft, it's likely the system can be turned off if deemed necessary, for example if you are avoiding a missile it might be less risky to dive close to the ground than to pull up.
I would guess there are two possibilities. The system won't override pilot input or the system only kicks in when it's 100% certain that a crash is inevitable.
> If the system predicts an imminent collision, an autonomous avoidance maneuver—a roll to wings-level and +5g pull—is commanded at the last instance to prevent ground impact.”
Sounds like by the time the system kicks in, you probably need it to kick in.
Engineering doesn't typically work this way. There needs to be a tolerance because you can't account for all variables. The question really is, at what probability of likelihood of a crash do you want to attempt to avert it?
Averting 99.5%+ of crashes probably eliminates flexibility beyond what is appropriate for a fighter pilot.
> “Auto-GCAS continuously compares a prediction of the aircraft’s trajectory against a terrain profile generated from onboard terrain elevation data. If the predicted trajectory touches the terrain profile, the automatic recovery is executed by the Auto GCAS autopilot. The automatic recovery maneuver consists of an abrupt roll-to-upright and a nominal 5-G pull until terrain clearance is assured.”
Any nose dive will look like it will hit the ground, won't it? The question is at which attitude does it need to pull up. Obviously, there are many factors involved, but there may be other factors the system is not aware of, that's why it probably makes sense that's overridable. I bet there is a way to do it.
It is mostly likely timing based. It may well calculate how much time is left until collision, or perhaps until a 5G pull will become insufficient to recover. When that time remaining drops below N seconds (which from looks to be perhaps 1-3 seconds if it is the until insufficient time metric, or perhaps an extra few seconds for time until collision) it will activate.
Obviously there are other criteria to prevent a landing approach from being seen as an imminent crash etc. But an override system that cannot be accidentally held by a disoriented or unconscious pilot seems plausible.
By the time it's 100% certain, there's nothing it can do. Any control input the system can make, the pilot could also be theoretically planning to make.
I didn't say "it's 100% certain that no matter what, there'll be a crash". Implied is that "100% certain of an impending crash without any extra input".
AGCAS was developed because pilots in the F16 specifically were blacking out from excess G forces and flying into terrain. An unconscious pilot isn't likely to make any useful control input at all.
Watching videos of "normal" people riding as passengers in fighter jets, you often see them pass out briefly. The pilots are have the training and strategies to push their G limits much much higher, but even so it's still a risk.
I wonder how long until we see a G-LOC accident in a modern fast accelerating electric car... I'm sure today's fastest accelerating street legal cars can do things to a body/brain that some people wouldn't be able to tolerate. (And also, if you consider how much ground you cover in the very brief 0-60 of a modern Tesla, it suggests that a poor choice with the right foot could lead to a very quick bad situation!)
Two things make that unlikely; cars have to grip the road and by the time you're getting close to G limits tires will start to skid and second the forces aren't in the right directions. For the former the closest a car gets to the levels that induce G-LOG are Formula 1 cars during braking (~5G) and during cornering (4-6G) [0] so even the highest performing cars out there you're just barely approaching where you might threaten G-LOC. The latter is why I think it won't really happen though because the acceleration is (basically [1]) never vertical like in fighter jets during climbs and turns. The forces in a car are, w.r.t. the human riding, forwards, backwards, and side to side and those directions don't force blood out of the head. Braking gets closest because the feet are furthest forwards but it's not quite the same.
[0] Acceleration is a measly ~2G at peak
[1] To avoid pedantry there is a case but that's on extremely high bank turns that only exist on test tracks really, NASCAR track banks might get close but I don't think get Gs there get high enough before cars break free up into the wall to threaten G-LOC.
Indy cars IIRC have run into the problem on banked tracks.
They did something to slow the cars down and remove the need for G-suits.
Indy cars got to around 5G, which is pretty hard to sustain without a G-suit.
The 9G stuff in the F-16/F-22 absolutely requires the G-suit and the reclined seat which changes the force vector.
I've motorcycled on a NASCAR track, but not a very steeply banked one relative to the superspeedways. Even the lower banking changes the way turns feel in a very dramatic way though.
That's for young, fit, etc people. Some quick googling says a particularly susceptible person might very well pass out at as low as 3 Gs. Still way below what a Tesla can pull, though.
It wouldn't necessarily need to be g-forces, given how many Americans have severe underlying health issues, you might just go over a rough bump and have a stroke or something.
Roller coasters tend to hit a good amount of gs, but if you pass out you'll be fine since you're not driving. One wonders if a Tesla detects its pilot unconscious can it bring itself to a full emergency stop. Automatically contact emts and then automatically distribute snacks to the arriving EMTs
> That suit dynamically responds to what the jet is doing, and squeezes the pilot’s lower body, like a high-tech blood-pressure cuff.
I’m generally bearish on the supposed incoming AI dark winter, but aircraft navigation and dogfighting seem like they’re going to be much better suited flown by a computer.
They can design aircraft that are much smaller and lighter, and can maneuver in ways that would cause a human to black out instantly.
You can fly a drone from your nice safe office which doesn't need to be anywhere near the war. No need for AI, though that is already used for the easy parts. For exactly the reasons you state.
Tom Scott (who has a great channel btw) went through a simulator for it too. The complete reboot of people passing out is really interesting.
For a funnier version (caused mostly be excitement rather than lack of blood) that doesn't involve people being moments from disaster check out the numerous Slingshot ride videos of people passing out.
Actually - I sort of want to work in defense for a few years to find out, that and the sheer amount of tech in a modern missile (the fact that they would likely be used is why I didn't apply) - I reckon most the algorithms here are basically finely tuned signal processing and other non-"AI" processes.
There's a tendency to think that everything that isn't completely ab initio like machine learning is bad or inelegant - I often fall for this trap myself, but to just get the job done you can get a very long way with "dumb" algorithms and a practically infinite budget.
Algorithms that don't use machine learning aren't dumb. The point of machine learning is you need to implement some function, can't derive it directly from first principles, but have lots of examples of correct inputs and outputs and can use that build a function estimator instead. But in a whole lot of quite sophisticated and intelligent applications, you can derive the functions from first principles. Not easily. One of the projects I spent the first few years of my career working on was the common image formation processing for spy satellite collections, and you need to take into account special relativity, orbital mechanics, curvature and velocity both of earth and of the arrays, midnight crossover in time keeping, temperature calibration and the impact on reported voltage detection of each sensor cell, parallax effects of rapid altitude changes, polarization of light. There's a ton that goes into it, but we know the physics and don't have to use statistics to make educated guesses. We can compute "given voltage levels x1, x2, ..., xn on sensor cells y1, y2, ..., yn at times t1, t2, ..., tn, that is what we were looking at" exactly.
Mind you, I'm talking basic level 1 transformation of raw data streams to human-intelligible images. Once you get into automated object recognition, that's when we start to use machine learning, but the algorithms upstream of that are still plenty smart.
As a theoretical physics student who can actually write good code, this is the kind of thing I want to work on - unfortunately where I live you basically have a choice of making things and making money
Part of it is that predictability is a desirable feature in these systems, but also that problems like the one described in the article you don't really need things like ML. The majority of control problems like this are surprisingly straightforward. They might be complex, in the sense of having a lot of variables, but the physics involved is well understood and can be modelled using traditional techniques.
The progress of self-driving cars is a good example of this. I can remember seeing expeimental self driving cars many years ago, but always going round mostly empty test tracks. Driving a car isn't that difficult for a computer system, what's hard is driving in highly complex urban environments with many other cars around that you need to predict.
Planes, in contrast, have a rather simple environment. The number of objects they have to avoid is massively lower, and their freedom of movement is higher, with established rules for how to behave, there are no traffic signs to interpret. This means that all you are really doing is object detection with radar, and collision avoidance.
In addition, modern combat planes are effectively flown by a computer all the time anyway, with the pilot providing the instructions. A number of fighter planes, especially the most modern, are essentially unflyable without computers due to their aerodynamics. Most are inherently unstable around at least one axis, which makes them more manouverable, but means they will not fly stably in the way a 747 will.
The advantage of the "old" control systems approach is that it's not a black box; you can reason about its behavior across the state space and make assertions about its frequency response and stability conditions.
(Is a neural network really different from a very large cascade of nonlinear filter elements?)
> Is a neural network really different from a very large cascade of nonlinear filter elements?
It's different in theory, but not by much practically. Reverse engineering the former is a fool's errand, but so would be a sufficiently complex version of the latter.
In terms of Kahneman's "Thinking Fast and Slow," ML, at least in its current stage, is like the fast thinking system. It's essential but is exponentially more valuable when combined with the slow system, which is still elusive in AI practice.
Indeed. I'm just kind of fascinated by military technology e.g. The F/A-18's cockpit HMDs and Hud were light-years ahead of anything in star wars with the exception of holograms even in 1983.
With a missile the actual signal processing and acquisition is what interests me rather than the control theory e.g. you send radar pulses out, you get Doppler shifted pulses back absolutely littered with ECM - now turn that signal into a target to aim at.
Laser codes on A2G missiles are a similar thing that interests me.
>> "the system consists of a set of complex collision avoidance and autonomous decision making algorithms..."
> It's 50 nested ifs, isn't it.
If that's the case, I don't see an issue if they're spread out among a reasonable number of functions.
I wouldn't want some sexy ML system that will sometime go off the rails because the clouds are in the wrong place to be within 100 feet of an aircraft control system.
Well, “it works” in your desktop testing. That doesn’t mean it works. How do you test the permutations of 50 nested if statements such that you know “it works”? Whoops, there’s a bug reported from the field. Have fun debugging that corner case.
Or IOW, why do you think the static analyser flagged it in the first place?
It doesn't matter if it's 50 nested ifs or not if we're talking about a recent generation jet fighter - it's going to be hell to test no matter what. Even with provable control systems you're still at the mercy of cutting edge technology and the whole point of the jet is superiority over a yet unknown adversary with high tech capabilities so the system has be overengineered with that in mind.
I dont really see a problem with that. While ugly as hell visually, it's stupid simple to find out what went wrong and fix it. Over abstraction and fancy technique doesn't automatically equate to reliability and performance. Why over complicate something just to inflate your own ego and be a show off to people who are not the end user putting their life in the hands of the software?
Assuming there are only “if” and no “else”, then there would be 51 code paths (Presuming other preconditions like: the condition is code or there is some code within each if; the code cannot throw exceptions; there is no return statement; etcetera).
You could perhaps have only two code paths: the 50 ifs are true and something happens, or any one of the ifs are false and something doesn’t happen... although that could be written as a single if statement with “and”s.
The 2^50 case would be more code than we can store (unless using techniques that reduce the complexity i.e. not 2^50).
I wouldn’t think it’s really that complicated. The autopilot on a 182 would do the same thing. If the nose is pointed at the ground, pull back on the yoke. Etc. The nicer ones can even stop a spin and have a button you press that’ll return you to straight and level flight. I guess a fighter is often purposefully in attitudes a regular plane normally isn’t, but I’d bet it’s a simple warning and override.
But I don’t fly fighters so I’m just guessing. Collision avoidance with other planes In a dogfight might be a lot more complex too.
Fighters sometimes fly close to the ground on purpose. Building a system that will allow extreme maneuvers near the ground without a false alarm trigger was hard. The Swedish air force was the first to use this system, and they routinely fly "close to the rock", through mountain passes. That's the hardest part of the problem, deciding when to initiate the recovery while not impeding what fighter pilots consider normal flight. The threshold is something like 0.3 secs before it's too late to correct. That requires a very good trajectory predictor for figuring out when it's going to be too late.
Next is getting the aircraft into a wings-level condition, no matter what direction it's pointing. For fighters, extreme attitudes are normal. Auto-GCAS will make violent maneuvers to do this. "At that instant, the Auto-GCAS commands some of the most aggressive, eye-watering maneuvers this ex-USAF flight test engineer and civil pilot has ever experienced.
If inverted (bank angle greater than 90 deg.) and somewhat nose-down, a negative 1g push throws the pilot "up" into his shoulder straps and lap belt to get the aircraft's nose headed skyward. Immediately, a 180-deg./sec. roll is commanded, bringing the aircraft to wings-level, right-side-up."
Only then can the system command a climb: "Somewhere after passing the 90-deg.-bank point, a 5g pull-up is initiated at an approximately 4g/sec. rate. The system commands a maximum angle-of-attack recovery, if flight conditions will not sustain a 5g pull-up."
The F18 literally has a switch labeled "Spin Recover". One of the things it does is to take the computer out of the equation and allow the pilot to fully command the airplane rather than processing stick and rudder inputs.[1] General procedure though is take your hands and feet off the controls to start with. The full procedure is in the link a few pages in. Notably, it ends with "if passing 10,000ft AGL with no indication of recovery, eject.
That is correct, but for boolean expressions, logical operators etc. it's just the one equal sign.
E.g.
if pilot.sleeping = true then alarm.playing := true;
// This will check if the property "sleeping" of pilot is true and will then set the property "playing" of alarm to true as well.
You could also do a "while" or "repeat .. until" there which would probably better to stop the alarm if the pilot wakes up again (... then alarm.playing := false).
It could totally work this way if you programmed a simple flight simulator with Delphi, even today.
BTW, one exception to := in assignments are initialized variables inside a var section:
sad that this didn't catch on more broadly. I find that the difference between `=` and `==` is consistently confusing to beginners. it's actually worse when someone has a strong math background. that and the fact that `1/2 * x` is zero in most languages...
Fully agree. This is why I still argue even today that (Object) Pascal aka. Delphi is a wonderful language and tool for beginners and casual programmers who want to get things done quickly and (somewhat) intuitively. Delphi unfortunately is not very en vogue at the moment due to misguided and community-destroying licensing policies by the company that owns it. But that's no fault of the language and the toolset. Both together make one the most mature, flexible and easily accessible environments for programming that I've ever encountered, from quick & simple UI-driven cross-platform apps to Web backends to hardware-oriented IoT stuff with a little Assembly sprinkled in.
Of course there is FreePascal and Lazarus and all that if you really want to play with Pascal without touching Delphi. Both are neat, even though Delphi is also now available as a free community edition.
That's because they glossed over the part that actually determines weather "concious" is true or false. They also skipped the content of flyto(base). But those may just be composition of many more simple things.
The art of programming is combining simple things to consisstently produce desired results during general usage.
I might be the only one, but reading the article and the discussion here, I can't help but ask a bigger picture question on whether it is worth putting so many pilots through the torture of flying F-16's and F-35's. G-locking sounds extremely unpleasant. Does it have any longer term effects on the pilot? Despite the development on the software, aircraft technology, and the fundamental aerodynamics side that the F-35 and F-16 programs have brought about, they still seem inhuman to the aviators.
Why isn't anyone questioning whether these high-maneuverability fighter aircraft should be used at all, especially given the huge budget of these programs?
How would they plan to fight against other high maneuverability aircraft without comparable or better aircraft? There is definitely a goal of going unmanned at some point in the future in order to save lives, but the job still needs to be done today.
And for those curious to see what an ADD activation is like, there's a video of a couple skydivers forgetting to pull their chute[0]. Those guys definitely got their lifes saved by it.
Question: What repercussions does a pilot face after an incident like this? Is there more physical training done do the breathing maneuvers, and plane training to prevent excess G’s like this? Is the pilot grounded for a certain period of time?
That was the most interesting bit to me. I am learning to fly an autogyro - 2000 to 3000 ft is our normal cruising altitude and I haven't been above 5000. The idea that you need to pull up at 4000 to not hit the ground says a lot about the speed.
Just to say, HNs had a comment yesterday with a terrifying video illustrating ‘the one big safety problem’ with autogyros - rotor disk unloading followed by “bunt over” (I think).
I hope you don’t mind me saying ... please don’t skip any safety training, particularly if you come from conventional aircraft.
I'm learning in the UK where you aren't permitted to skip anything. Even if you had a PPL for fixed wing (which I don't) it's only worth 15 hours of the required 45 in the gyro.
I'd be curious to see the video, but modern gyros are less susceptible to it and it's simple to avoid as a pilot (far easier than stalls and spins).
I find it alarming the degree to which US pilots can fly gyros with minimal specific training - they're very different to planes.
Warning: From the description: “This is a pretty sobering piece of film and its pretty obvious the pilot didn’t walk away. So if this is likely to upset you please don’t watch.”
Assuming we're talking about the typical example (https://youtu.be/WkZGL7RQBVw) - in the HUD, altitude is the right-hand vertical tape. (The left one is airspeed.) It looks to bottom out around 4400 feet during the AGCAS recovery.
I'd need to look it up to be sure, but I would assume by default that this is a radar, not a barometric, altimeter.
(To expand on that: A barometric altimeter works via air pressure, and thus shows height above sea level; a radar altimeter measures the time taken for microwaves to go from it to the ground and back, and thus shows height above ground. I wouldn't be surprised to find both types in a fighter, but I would be very surprised to find that a radar altimeter wasn't the default, because it's going to be the one that provides the most accurate information and thus the most useful to a pilot who needs to worry about avoiding CFIT during complex maneuvers.)
The radar altitude is shown further down the right side (the box with the "R" next to it). It bottoms at 2970 as I saw it at with the tape showing 4370MSL.
The right side of the tape is MSL.
If you watch that box through the video, you can see why you don't use the radar altimeter for everything. Any time you don't have clear line of sight to the ground with the belly of the aircraft, the radar altimeter blanks out completely. If you're in a roll, it will give incorrect information as it's not pointed directly at the ground.
1. I'm curious what happened to the pilots: did they remain in their role as fighter pilots or were they transferred to other airframes that pull fewer Gs?
2. I can't help but think about the possibility that software that can take over when it thinks the pilot is non-responsive could be hacked to crash on purpose or be taken over remotely. Perhaps this is a "feature" that will allow the F-16 fleet to be used as UCAVs without telling the public (or our enemies) about it?
I don't understand your second point. How could a F-16 be hacked remotely when there's no remote access? There's no attack surface. How is this different from any other fly-by-wire system?
There is remote access. Fighters routinely exchange data via datalink like LINK 16 and JTIDs. It's not inconceivable that this could lead to some exposure. It would depend on how segmented the flight control computers are from the other computers. A lot of the cyber warfare plans are attempting this type of attack.
Link-16 is JTIDS. It doesn't have any connection to flight controls or any controls of anything it's deployed in (which is just about everything). It's also type-1 encrypted.
If you're going to park 600 military planes wingtip to wingtip, you might as well just program the software to purposely crash them all into the ground. Didn't we learn anything from Pearl Harbor?
https://www.youtube.com/watch?v=WkZGL7RQBVw
That automated recover voice is quite eerie.