When Will We Have Unmanned Commercial Airliners?
spectrum.ieee.org
spectrum.ieee.org
I know that aircraft carriers have these systems, but having been on two deployments on an aircraft carrier I have never heard of it being used. I think it's only for emergencies when the pilot is injured or incapacitated.
Did you ever hear of something going wrong with the automation?
Interesting. So it hasn't disengaged at all, it's just decided to abort the current landing attempt, and go around to try again.
The author states that even in easy adaptations such as traffic watch we still do not see UAVs and what the author is missing is the cost! Those military UAVs cost almost as much as a private jet in what today using 1950s technology can be achieved for the cost of a used cluncker.
Also, the same inheritable problems that we face fully automating cars, exist in fully automating airplanes. Each is an individual unit and very autonomous with no single 'track' to follow. In addition to this largely autonomous environment you have integration problems, where you would integrate the old with the new automated environment and still be able to offer the same capacity of airspace usage.
As frustrating (and yes it is for pilots too - more than passengers realize) as air travel seems now days. The achievements of airspace utilization in places like New York, Chicago, Los Angeles, Phoenix, San Fran etc is almost mind boggling considering it's used with a mosaic of 1930s radar equipment which honestly can't pin point you to an exact location within a circle of 1 mile around yourself.
It's going to be a very, very long time before we see autonomous airliners. I'll talk about specific technical hurdles, but I think the biggest issue is psychological: it's one thing to entrust a bunch of freight to an autonomous vehicle, another thing entirely to entrust dozens of living, breathing human beings to such (the article discusses this, including the concept of "shared fate"). I am confident that autonomous airliners will only come into service when autonomous aircraft technology reaches a point where the computers are able to handle every aspect of flight safety better than humans. Right now they can already do some of those things better than humans, and those tasks have already largely been shed by human pilots and entrusted to their computers. I think there will be a gradual transition as the computers are able to take on more and more of the tasks. The article also mentions how this process is already causing basic aviation skills to atrophy in pilots who leave too much to the computers. I think this is a very real problem, and I think it was a major contributor to the Air France flight 447 crash. Prudent pilots do more manual flying than is strictly necessary because that's the only way to maintain proficiency. If this problem becomes severe enough, expect to see the FAA establish more granular proficiency requirements.
The author talks a lot about how much the military is using UAVs, and seems to think that this is a good model for civilian applications. It isn't. The military has an entirely different set of risk considerations than civil aviation. Take the example of medivac UAVs: a medivac UAV will almost certainly have a higher mishap rate than a manned medivac helo, which would be completely unacceptable for civilian purposes. However, for the military that increased mishap risk is more than offset by the risks of putting an entire human crew into harm's way just to medivac a single wounded soldier.
Military UAVs currently in use generally have much higher mishap rates than their manned counterparts, but the military tolerates this because aicrew don't die in UAV mishaps, and UAVs are generally less expensive to replace than manned vehicles. Part of the reason for this is that features designed to prevent or mitigate mishaps cost money, and it is often cheaper to leave many of them out and accept the higher mishap rate, especially when no human crew is involved. However, part of it is that autonomous systems still just aren't as good at flying safely. For the military, the benefits outweigh the costs, but I really can't see a for-profit corporation reachig the same conclusion.
>Northrop Grumman has built some sense-and-avoid savvy into the unmanned helicopters and other UAVs it's developing for the U.S. Navy.
I happen to be intimitely familiar with one such system, and somewhat familiar with another. This sentence is utter BS. I can only assume that the author was fed a line by an NGC PR-type and took it at face value. A more honest way to say it would be:
Northrop Grumman is trying to incorporate limited sense-and-avoid capabilities into the unmanned helicopters and other UAVs it's developing for the U.S. Navy.
Moreover, FAA has a requirement for "see and avoid," not "sense and avoid." The military is trying hard to sell them on the idea that it should be "sense and avoid," of which "see and avoid" would be just a subset, but so far the FAA has remained deeply skeptical. The FAA is right to be conservative about this change: none of the currently proposed systems would be as effective at collision avoidance as the Mark I Eyeball, and thus far the systems I am aware of (to which the quote from the article was referring) are still a very long way from working properly. This seems to me like the kind of technical problem that eventually can be overcome, but it's going to take a lot of work to make that happen.
Because "see and avoid" is so critical to safety of flight, and because UAVs can't currently do it, the FAA does not allow UAVs to operate in its airspace, with some tightly restricted exceptions for military and law enforcement UAVs. I doubt very highly if they would make similar exceptions for civilian purposes, and even if they did, the restrictions involved are so limiting that there aren't many viable applications.
I imagine with sufficient smart people working on it, flying airplanes in relatively uncluttered sky would yield results faster.
As for see-and-avoid, perhaps it works for obstacles on the ground, but other aircraft are moving so fast, is it really feasible to eyeball them before they are on you? Perhaps in pursuit, but at any significant angle they flash past at hundreds of miles an hour. Only radar etc has a chance of identifying/avoiding at those speeds. <Edit: spelling>
Having experienced plenty of both, I can tell you that this is utterly false.
You could equally say adding up big numbers is much harder for you than walking across the street. To a computer the former is trivial, the latter no walking robot has yet done.
I think that two big reasons for this are probably:
1: For land vehicles, simply stopping in place is almost always an effective collision avoidance tactic (unless the other vehicle is deliberately seeking a collision). This simple solution is not available to airplanes.
2: Tracking objects thar are moving in three dimensions with a sensor that is also moving in three dimensions is an immensely more complex problem than tracking objects that are constrained to move on a fixed surface in two dimensions with a sensor that is also constrained to move on a fixed surface in two dimensions.
My point was, there is another wrinkle. The land-nav problem was opened up, made a competition with a big marketing budget. It was also intractable, unsolved, too hard. Until lots of smart people started brainstorming and trying crazy things and cooperating.
Airplanes can change speed drastically, which is at high speeds about as effective as stopping. And no, you don't get to say collisions are hard to avoid because 3 dimensions are hard to calculate, making that not a solution.
I think I begin to see why the problem hasn't been solved yet.
By your logic, every "impossible" problem could be solved easily if just DARPA would offer a small prize to whoever solves it. Unfortunately, the real world doesn't work that way. There's a reason why DARPA chooses the tasks they do for their challenges: they spend a lot of time and effort identifying tasks that are highly likely to be amenable to novel solutions.
>Airplanes can change speed drastically, which is at high speeds about as effective as stopping.
Incorrect, on two counts. First, not all airplanes can change speed "drastically." Second, it is not as effective at preventing a collision as stopping. If both cars in an impending collision stop (and in many cases, even if only one of them stops), a collision becomes impossible. On the other hand, there are a lot of situations where deceleration merely delays, but does not prevent collision. That has value, but it's not as good.
>And no, you don't get to say collisions are hard to avoid because 3 dimensions are hard to calculate, making that not a solution.
I never said it was "not a solution," but I definitely do get to say that it's a much harder problem to solve. Here's the steps you have to perform to avoid a collision:
1: Detect an object.
2: Track the object to determine it's course and speed.
3: Compare the object's course and speed to yours to determine how likely a collision is.
4: If the probability of a collision is unacceptably high, determine a change of course and/or speed which will reduce the probability of collision to an acceptable level. If the probability of collision is already acceptably low, return to step 2.
5: Maneuver to change course and speed accordingly, then return to step 2.
If you are moving in three dimensions and the objects with which you might collide are moving in three dimensions, step 2 is hard to do accurately. (Unless you've studied radar tracking, you probably don't appreciate exactly how hard, but if you're genuinely interested, Skolnik's Radar Handbook is a good place to start.) The less accurate you are at step 2, the harder steps 3, 4, and 5 become, because you have to deal with more uncertainty. Is that really where the other object is? Is that really where it will be in twenty seconds? How certain are you of that? How certain are you that there really is something even there at all? If you're wrong in one direction, you'll have a midair; if you're wrong in the other direction, you'll perform some extreme maneuver for absolutely no good reason.
I imagine with sufficient smart people working on it, flying airplanes in relatively uncluttered sky would yield results faster.
As I mentioned, many of the tasks of flying airplanes have been successfully automated. However, there are significant complexities that pilots must deal with that don't apply to cars, and most of those still require human intervention, especially in-flight emergencies. If your car engine quits, you pull over to the shoulder, turn on your hazard lights, and call AAA. If your aircraft engine quits, the response is much more complex. If the power steering on your car quits, you do pretty much the same thing as described above. If your flight controls malfunction in an airplane, the response is much more complex. If your car catches on fire, you do the same as above, plus get out. If your airplane catches on fire, you've got a much bigger problem. I could go on, but I think that gets the idea across.
Also, the sky is not "relatively uncluttered." There are a lot of airplanes flying at any given moment, and most of them are concentrated onto airways. It gets even worse in the terminal area: lots and lots of planes coming and going to and from many different directions, all in a very small piece of sky.
>As for see-and-avoid, perhaps it works for obstacles on the ground, but other aircraft are moving so fast, is it really feasile to eyeball them before they are on you? Perhaps in pursuit, but at any significant angle they flash past at hundreds of miles an hour. Only radar etc has a chance of identifying/avoiding at those speeds.
I am alive today because, on countless occasions, I and my fellow aviators have looked outside, seen another aircraft, and maneuvered to avoid a potential collision. I think that you really don't have an accurate mental image of how this works, so I'll try to explain a little bit:
Consider two airliners cruising at 424 kts each, one heading due West, the other due North. The rate of closure is 600 kts. Depending on the atmospheric conditions, they will be visible to each other at about 10 nautical miles, which gives them an entire minute to spot each other and maneuver to avoid a collision. Even if it's two sueprsonic fighters flying right at each other at 600 kts each, they still have thirty seconds to spot each other. A much more realistic scenario would involve two aircraft in the terminal area, where they would be moving much more slowly, giving them much more time to see each other and respond.
I think the most dangerous situations are where two aircraft are on headings that are different by less than forty-five degrees: they are basically next to each other, closing from each other's sides where they are less likely to be spotted. The rate of closure is probably very low, but that's more than made up for by the awkward situation in regards to field of view from many cockpits.
I think that unmanned plane has actually a much better chance of surviving fire, it could have an inert atmosphere, or it could be unpressurized so fires would be much less frequent. Also fires happen mostly on freight planes (lately UPS and Asiana) and freight planes would be probably easier to certify for unmanned flying than passenger planes.
> I am alive today because, on countless occasions, I and my fellow aviators have looked outside, seen another aircraft, and maneuvered to avoid a potential collision.
Isn't this what Traffic collision avoidance system (TCAS) is for? The pilots already have to do as they are told by TCAS (after the collision over Switzerland). Surely managing traffic of obedient agents with known limits of performance isn't that hard - if all the planes were unmanned there wouldn't be problems
However, the points that you mentioned about other emergencies and passenger traffic still stand.
This would be very costly to implement, and probably very heavy. That being said, inert gases are used in places where arcing is likely (e.g. radar waveguides).
>or it could be unpressurized so fires would be much less frequent
I really don't think this would have a significant impact on the frequency of fires. Plus, a lot of avionics need particular environmental conditions, in some cases including pressurization.
>Also fires happen mostly on freight planes
High-power electronics, such as military radars, also pose an increased risk of fire.
>freight planes would be probably easier to certify for unmanned flying than passenger planes.
If you only cared about the contents of the plane, this would be true. But what happens when a flaming ball of wreckage that used to be an unmanned freight plane plows into a suburban neighborhood, a school, or a downtown skyscraper?
>Isn't this what Traffic collision avoidance system (TCAS) is for? The pilots already have to do as they are told by TCAS (after the collision over Switzerland).
TCAS isn't all it's cracked up to be. First of all, it only works if the other plane has a transponder (there are still plenty of light civil aircraft out there with no transponders). Second, there are different versions out there with different levels of accuracy. The older kind are not very accurate at all, and basically serve only to give the pilot a general idea of where to look in order to spot the traffic and avoid the collision the old-fashioned way. The more accurate kind only works if both aircraft involved have the necessary equipment.
As for the collision over Switzerland, the reason for that rule is that the collision happened in part because TCAS and ATC gave conflicting instructions: one told plane A to go up and plane B to go down, the other told plane A to go down and plane B to go up. One crew did what TCAS said, the other did what ATC said, and they both ended up descending. So the reason for this rule isn't that TCAS is a magical panacea for midairs, but rather a way to consistently resolve any future such conflicts between TCAS and ATC.
>Surely managing traffic of obedient agents with known limits of performance isn't that hard - if all the planes were unmanned there wouldn't be problems
If all the planes were unmanned, then managing traffic would be much easier, but that still leaves other issues, like where a plane ends up when it malfunctions and crashes. It would also require a wholesale changeover that simply isn't plausible.
Why hasn't it been made mandatory that all aircrafts should have a standardized transponder ? It scares me a bit that, at the end of the day, we rely on pilots avoiding collisions by sight.
In some parts of the world, it may be. For example, I don't know if Europe allows for aircraft without transponders. Even in the U.S. you must have a transponder to enter certain types of airspace (e.g. the airspace around major airports). As far as I know, no matter where you are in the world, you must have a transponder to fly IFR.
Installing a transponder is a non-trivial expense, especially in older aircraft (which are the aircraft most likely to not have transponders). For a lot of small aircraft in the U.S. this would represent an unnecessary burden on the owners. For example, crop dusters: they typically fly around low and slow in areas with very little traffic, under day VFR conditions, so they have no need to interact with ATC and therefore no real use for a transponder.
>It scares me a bit that, at the end of the day, we rely on pilots avoiding collisions by sight.
We don't rely solely on this: we have transponders (with TCAS in some cases), ATC radar, and (in some cases) airborne radar. All of these tools help us to avoid collisions. Unfortunately, none of them are 100% effective, and in most of the situations where they all fail, the good old Mark I Eyeball usually saves the day. See-and-avoid isn't perfect, either (if it was, we'd never have midairs), but it is still the most effective tool available for avoiding an impending collision.
Unfortunately regulations and liability costs make anything in aviation expensive. So most small planes are running around without tcas.
(If your plane engine quits - what would a human do? whatever it is, a computer could be programmed to as well. Near collisions - a constantly vigilant computer vision system watching out the window is more likely to avoid collision than a pilot with 30 seconds of reaction time.)
This depends on a lot of factors. Some of them are things that a computer can probably be programmed to consider properly (e.g. specific cause of failure). Other factors require judgment, such as your more general situation: depending on where you are and what else is going wrong, you might chose to land the airplane at the nearest appropriate airfield, or you might choose to continue on to an airfield farther away where you can get better support while attempting to restart the failed engine on your way there, or you might decide that you have to get it on the ground right now, and that empty farmer's field over there looks good enough.
In case of more catastrophic failures like fire, computers become even more problematic because the sensors they depend on for inputs can be damaged or destroyed, leaving them with insufficient information to act properly.
>Near collisions - a constantly vigilant computer vision system watching out the window is more likely to avoid collision than a pilot with 30 seconds of reaction time.
People with a lot of money and resources have been trying to develop a fully autonomous system for avoiding impending collisions. They will almost certainly eventually succeed, but so far they haven't even come close to being as visual scan by human pilots.
[EDIT: resolved ambiguous use of "field.")
I believe the technology to solve the problem is out there, it's just a matter of the right people trying at it.
Computer vision is close to being solved. Look at kinect, kinect 2/google goggles. People inside google/microsoft are racing at this. I'm sorry, i have to disagree with your pessimistic attitude on this.
With regard to fire - fire can kill human pilots too. With sensors, you can create a multiply redundant system - put in 20 extra sensors. With humans it's not possible.
A billion dollars was spent by the eu in the 80s on self-driving cars. They didn't completely succeed. It looked like noone would succeed for 30 years. And yet, bang, when the competition's opened up, a couple of guys from stanford do it. I believe the technology to solve the problem is out there, it's just a matter of the right people trying at it.
If they tried in the '80s and the guys from Stanford did it in the 2000's, then it was almost 30 years before anyone succeeded. I think that success had a lot more to do with technology maturing over time than it did with "the right people trying at it."
>Computer vision is close to being solved. Look at kinect, kinect 2/google goggles. People inside google/microsoft are racing at this.
This really depends on what you mean by "solved." Kinect is a hell of a long way from what you would need to avoid collisions in a 3D space. Kinect basically just has to deal with the outlines of objects at a relatively narrow set of distances. When your sensor is moving in three dimensions and you are trying to track an object that is also moving in three dimensions it gets a heck of a lot harder, even if you are using radar (which gives you range). If you're trying to figure out range based on the apparent size of an object of unknown actual size, it gets even harder.
>I'm sorry, i have to disagree with your pessimistic attitude on this.
I'm actually quite optimistic that it will happen, just not for many years yet.
>With regard to fire - fire can kill human pilots too. With sensors, you can create a multiply redundant system - put in 20 extra sensors. With humans it's not possible.
If your engine is out on a wing and it catches fire, the fire sensors will tell you so, and shortly thereafter they will most likely be destroyed and tell you nothing further. An engine on fire out on the wing is not going to burn up the pilot. The pilot can look out the window and quickly and easily assess the condition of the engine and the wing: did it burn out, or is it raging out of control, or maybe there are subtle signs that indicate something in-between? Maybe with enough fire sensors scattered all over the plane a computer could make a similar assessment, but you're talking about a lot of extra money and weight, and you still have the problem that your sensors are going to burn up shortly after going off and then you have no idea if the fire has gone away or if it has just stopped spreading. Someday maybe you can give the computer a camera to "look" at the wing to make the same kind of assessment that a human pilot can make, but that is a very long way off.
- computer vision problems (e.g. plane catches fire, how do you tell how much fire etc.)
- tracking other objects in 3d while in moving in 3d at high speed
The question is - can humans do this? If yes, computers can do it eventually. The only question is how long away is this. What we know from machine learning, is that data is important. If you can gather enough data you can do anything. So really, your problems are a question of data collection. It is not a technically difficult problem. (By the way, one of the problems you have in aerospace is, is that you're control theory heavy rather than pro-ai, which means you end up not being able to solve the difficult problems.)
Also, the eu 80s project ended around early 90s, and the grandle challenge win was only 15 years later, not 30. Had the challenge been tried 5 years earlier, it would have worked. The algorithms and hardware was already sufficient.
>The question is - can humans do this? If yes, computers can do it eventually.
This logic is deeply, deeply flawed. I happen to believe that computers can eventually perform the tasks under discussion, but "humans can do it" is not one of the reasons why I believe that.
>So really, your problems are a question of data collection. It is not a technically difficult problem.
When it comes to tracking airborne targets with airborne radar, data collection actually is a technically difficult problem. The combination of waveform, antenna design, transmitter design, receiver design, tracker design, etc. present a set of engineering tradeoffs in effective range, range resolution, azimuth resolution, weight, size, flase positive and false negative rates on radar returns, and other performance characteristics. Even the very best airborne radars provide data which is limited, especially in terms of accuracy and precision.
>By the way, one of the problems you have in aerospace is, is that you're control theory heavy rather than pro-ai, which means you end up not being able to solve the difficult problems.
A little more about my background: my BS is in EE, with a specialization in microcomputer interfacing (I took a lot of CS classes). In grad school, my stability and control prof had actually done some pioneering work in incorporating non-linear logic into stability and control systems (don't try this at home, kids). In addition to stability and control, my other focus for my MS was avionics. The prof who taught most of my avionics classes was actually from the CS department (his undergraduate background was EE, with a specialization in radar). One of the things that kind of surprised me about aero, having come from EE, was how broad the discipline is. Before going back to school for aero, I thought that getting a degree in aeronautical engineering would be primarily about aerodynamics, with a smattering of other stuff. Instead, I discovered that everyone gets a little bit of everything (aerodynamics, propulsion, structures, stability and control, avionics), and then specializes in one or two particular areas. By the PhD level, people who have specialized in areas other than aerodynamics have largely forgotten most of what they learned about it as undergrads. It's an incredibly heterogeneous field: propulsion and structures guys have more in common with MechEs than with other AeroEs; avionics and stability and control guys have more in common with EEs than with other AeroEs; etc. So to characterize the discipline, or any individual within it, as "control theory heavy rather than pro-ai," displays a deep misunderstanding of the character of the community. I guarantee you that there are plenty of AI experts working in the aero field.
>Also, the eu 80s project ended around early 90s, and the grandle challenge win was only 15 years later, not 30. Had the challenge been tried 5 years earlier, it would have worked. The algorithms and hardware was already sufficient.
This just reinforces my broader point: success came not as a result of some innovative genius applying a novel new approach but rather because the technology had matured--over the course of several years--to the point where success had become not only possible but likely. Radar tracking is currently experiencing big advances for that same reason. The theory behind Space-Time Adaptive Processing (STAP) has been around for decades, but the available technology has not been up to the task of implementing it effectively. In the past we've resorted to less effective tracking methods such as MTI, but in the last decade or so the technology has finally made STAP reasonable to implement.
I will offer another example. Why did the aerospace dudes not be able to autonomously fly a helicopter. in 2004, andrew ng decided to tackle this. He completely ignored any previous work, just using a dumb algo (reinforcement learning) and laptop managed to get amazing autonomous performance out it. Why was it him (ai researcher), and not people from the field of flying.
Not especially. I audited a course as an undergrad, and hardly remember anything from it now, but I have a layman's understanding of the basics.
>There was a debate about whether "hand engineering" vs "dumb simple algo's" would get results. Dumb algo's won.
There are design tasks for which "algorithms" are better suited, and there are design tasks where experienced human engineers still do far, far better. The statement "Dumb algo's won" is certainly true for some applications, but not all.
>Your mentioning of MTI and STAP, and how difficult radar design is etc. etc.,
Now it's my turn: you clearly are not well versed in aviation or in radar principles. Not every problem can be magically solved by throwing AI at it. Radar theory is well established, and the equations are well known. Unfortunately, they are hard equations to solve: determining the location of an object using radar involves some complicated math with a lot of variables, and the only way to solve those equations is to chew your way through them. You can simplify them, but then you have to accept increased errors from the terms you throw out.
Where "algorithms" come into play is tracking, and depending on how you define AI, radar engineers have been using AI since the invention of the first automated tracker. Even the very best automated trackers in existance today are not nearly as good as an experienced operator looking at raw returns. Someday that will probably change, but that day is still many years away.
>...makes me think you aero guys are still in hand-engineering land.
As I said before, you're making a huge mistake by lumping "you aero guys" into a single group. "Aeronautical engineering" is really "every other kind of engineering, applied to aviation." When I was in grad school, one of my buddies' thesis was pure AI: he developed a learning algorithm for choosing the optimum path for a jet to taxi around a crowded flight deck, using DGPS as the only position source. Another guy combined machine learning with CFD in an attempt to design better supersonic lifting surfaces (the results were not good, but his thesis was still a "success" in the sense that he expanded human knowledge and the general concept showed promise). There are some applications where AI is the way to go, and there are some applications where what you derisively call "hand engineering" is infinitely superior.
>I will offer another example. Why did the aerospace dudes not be able to autonomously fly a helicopter. in 2004, andrew ng decided to tackle this. He completely ignored any previous work, just using a dumb algo (reinforcement learning) and laptop managed to get amazing autonomous performance out it. Why was it him (ai researcher), and not people from the field of flying.
Time for a history lesson:
The Navy deployed the first effective autonomous rotorcraft in the 1960s: http://en.wikipedia.org/wiki/Drone_Anti-Submarine_Helicopter
So the "aerospace dudes" were "able to autonomously fly a helicopter" before Andrew Ng was born, and they did it using "hand engineering."
Not autonomous enough for you? Firescout flew autonomously four years before Andrew Ng flew his helicotper autonomously: http://en.wikipedia.org/wiki/Firescout
Firescout was also developed by "people from the field of flying."
Your point about complicated mathematical equations lies at the root problem of you "unified engineering" guys. modern ai (machine learning) is where you give up on the assumption that you (puny human) can impart "wisdom" to your system. You simply throw a random set of equations (a neural network) that are large enough/not too large (overfitting) to capture physical reality. Getting the errors low is a matter of getting enough data and experimentally adjusting the size of your nnet.
Yes, there mught be grad students and profs trying ai to solve aero problems, however, if enough resources are not devoted, they will not yield good enough results. For example, spend a billion dollars (gathering data/computation) to solve your radar problem. A billion dollars in your field is pocket change.
You have absolutely no idea what the performance and handling characteristics of Firescout are; in fact, it is apparent that you lack the domain knowledge to understand their meaning even if they were presented to you. Nevertheless, you assert without hesitation that Andrew Ng's helicopter is superior. This is the epitome of fanboyism.
>Your point about complicated mathematical equations lies at the root problem of you "unified engineering" guys.
You clearly don't know what you're talking about here. Go read Skolnik's Radar Handbook, then try to tell me with a straight face that random processes are going to derive those equations for you, and somehow magically come up with a way to sidestep the basic reality that they have to be solved.
>modern ai (machine learning) is where you give up on the assumption that you (puny human) can impart "wisdom" to your system. You simply throw a random set of equations (a neural network) that are large enough/not too large (overfitting) to capture physical reality. Getting the errors low is a matter of getting enough data and experimentally adjusting the size of your nnet. Yes, there mught be grad students and profs trying ai to solve aero problems, however, if enough resources are not devoted, they will not yield good enough results. For example, spend a billion dollars (gathering data/computation) to solve your radar problem. A billion dollars in your field is pocket change.
In order to use a tool effectively, you have to understand both it's capabilities and it's limitations. Even though you have indicated that you are an expert on machine learning and I have admitted that I am not, it is now abundantly clear to me that you have absolutely no understanding of the limitations of machine learning.
If only every problem could be solved optimally by simply throwing enough data and a big enough net at it. Unfortunately, that's not how the real world works.
One problem is that machine learning algorithms often converge on local maximums that are far less optimal than is possible. The guy who worked on the taxi routes had enormous issues with this, and only after extensive tweaking was he able to come up with solutions that were on par with human path-choosing.
An even bigger problem stems from the fact that a machine learning algorithm is only as good as the model it works in. I mentioned that the guy working on lifting surfaces in CFD did not get great results. His problem was that his algorithms pretty much always found the places where the CFD models diverged from reality: they would find the optimum shapes for the model they were working within, but those shapes always performed terribly in the wind tunnel because the algorithm was finding optimums at points where the model diverged significantly from reality. You can't solve this problem with "better models," because every model diverges from reality. If a model doesn't diverge from reality, it's no longer a model, it's reality. Where he really impressed his review board was when he detailed a follow-on experiment of using this phenomenon to develop better CFD models, within which human engineers would be able to come up with better designs.
Finally, a billion dollars is not "pocket change" in any field. Even if someone had a spare billion dollars laying around to fund R&D for radar tracking, the opportunity costs of blowing it on a machine learning experiment, instead of using to fund experienced engineers working from proven principles, would be unacceptably high.
Of course, dr andrew ng, head of the stanford ai lab is pussying around with his autonomous helicopter, after all problems were apparently solved by your defence contractor buddies in the 60s. The fact that helicopter pilots still exist is because society is too rich and we need to lighten our wallets. There, i mirrored your appeal to authority argument.
Mirroring your insults, it's clear you don't know what you're talking about. The fact that you've somehow been granted a doctorate further confirms my suspicions about the quality of education these days. The simplicity/non-pioneering-ness of your phd buddies's theses' is further confirmation of that. And the fact that you have been assigned to evaluate important technologies in your sector says a lot about the general competence level in it.
Of course searching can result in local minima. -That Exactly- is why you have to keep to keep running computers and getting more data. You can keep chanting to yourself - i am clever, i am clever, i write equations - and tell everyone the problem is difficult, years out from solution - or you can switch on the damn computers and let them find your answer.
If a billion dollars is too much for a system that can finally allow you to have autonomous planes, that you hope somehow your big brains will solve it, despite not having done so for a few decades, means that you, or your industry, does not have a clear grasp of the meaning of the term opportunity cost.
I did not cite the textbook as an appeal to authority: I cited it because you repeatedly demonstrated that you don't understand what I'm saying, and kept making ludicrous arguments as a result, and reading that book (or a similar one) would be the only way for you to gain the necessary domain knowledge in order to say something meaningful on this subject.
Similarly, I raised the issues of my peer's thesis work not as an "appeal to authority," but as a concrete example of the limitations of machine learning as an engineering tool.
You, on the other hand, used Andrew Ng's repeatedly as an appeal to authority. The worst part is that your primary example was factually incorrect: you initially stated that he was some kind of wunder-kind who was able to easily solve a problem that had supposedly been impossible for regular aero engineers to solve; when I pointed out that regular aero engineers had, in fact, solved the problem two decades before his birth, you responded with the absurd claim that his work was somehow superior, despite a complete lack of evidence to support that position.
Moreover, saying, "you do not know what you are talking about on this subject" is not an insult. I tried saying it more subtly at first, with attempts to fill some of the gaps in your domain knowledge, and yet you persisted in making arguments based on terribly insufficient knowledge of the subject under discussion, so I came out and said it explicitly. When I did so, I even provided a text you could read in order to correct your ignorance, but you chose to reject that as "appeal to authority."
>The fact that you've somehow been granted a doctorate further confirms my suspicions about the quality of education these days.
I never claimed to have a PhD. I have clearly stated that I have a MS in Aero.
>The simplicity/non-pioneering-ness of your phd buddies's theses' is further confirmation of that.
Just as you claimed that Andrew Ng's helicopter was somehow superior to other autonomous helicopters, even though you know nothing about those other helicopters, you now claim that the graduate thesis of two complete strangers are "simple" and "non-pioneering" based on a few sentences I wrote. Throughout this conversation, you have displayed this habit of reaching unreasonable conclusions based on insufficient evidence. Your arguments would be much more plausible of you would get rid of this habit.
>Of course searching can result in local minima. -That Exactly- is why you have to keep to keep running computers and getting more data. You can keep chanting to yourself - i am clever, i am clever, i write equations - and tell everyone the problem is difficult, years out from solution - or you can switch on the damn computers and let them find your answer.
This sums up the fundamental problem with your views. I have stated repeatedly that machine learning has its uses, but that it also has its limits, and that many aspects of engineering and design are still best conducted by human beings. I have given several examples to demonstrate this. You have this inexplicable faith that any problem can be solved just by throwing enough data and computers at it. It would be wonderful if only all engineering problems were that easy to solve. Unfortunately, it's just not true. If it were true, people would be disrupting the industry en masse by having computers design superior products faster and cheaper than human engineers can. You even add a touch of "No True Scotsman" to your reasoning: if you don't get magical results from your machine learning, it must be because you're doing it wrong: not enough data, or not enough computers, or you didn't spend enough time tweaking it; just throw more time and money at it, and then you'll get the answer.
Yes, you are correct, I have come to a viewpoint that all problems can be solved by throwing enough computers and data at it. This was informed by arguments in ai. See for example, jurgen schmidhuber's website, or genetic programming at john koza's website.
I agree, many problems are in a sense "easy", and human's can solve them. However, my belief is that those are the problems that have already been solved. The difficult problems, the ones that have not yet been solved, might well be too difficult for humans to comprehend. Computation is cheap enough now that it should be the default first step to try to brute force a solution. Even in high school, teach students how to describe problems as an optimization. Don't bother teaching them equation solving. Analytical solutions are sometimes needed, let that be a specialization for advanced undergraduates or even graduate school.
This is kind of like the reductionism vs non-reductionism argument. In physics simple laws were discovered, however, in biology this will not be possible.
An I apologize as well, as my statements clearly could have been made in a more conciliatory tone.
>Yes, you are correct, I have come to a viewpoint that all problems can be solved by throwing enough computers and data at it.
This may eventually become true, but we still have a lot of progress before we get there. I suspect that when it does become true, computers will look a lot more like animal brains than the computers of today, or maybe they will look like something completely different from either.
>This was informed by arguments in ai. See for example, jurgen schmidhuber's website, or genetic programming at john koza's website.
After a very cursory look, it appears that Dr. Koza has a very pragmatic attitude about genetic algorithms and is well aware of their limitations, and thus choses to focus his efforts on areas where they are most applicable. On the other hand, Dr. Schmidhuber seems to have staked his legacy on the idea that computers will soon be able to solve absolutely any problem better than humans can, and is passionately trying to spread this vision. He may very well be proven correct in the end, but I tend to be deeply skeptical of predictions made by such visionaries.
>I agree, many problems are in a sense "easy", and human's can solve them.
Some problems that are easy for humans are hard for computers, and some problems that are easy for computers are hard for humans. I assume that's why you put "easy" in quotes: problems that are "easy" for humans.
>However, my belief is that those are the problems that have already been solved. The difficult problems, the ones that have not yet been solved, might well be too difficult for humans to comprehend.
We may very well eventually reach a point where we truly have solved all of the problems within our capacity as humans, but that day is so far off that everyone alive today will be ancient history by then. Often, solving one problem reveals several more interesting problems that we hadn't even considered before.
>Computation is cheap enough now that it should be the default first step to try to brute force a solution.
This may be true for some classes of problems, but it is still not true for many, and will never be true for some. Consider cryptography: "brute force" only works if you have considerably more computation power than the computer(s) used to perform the encryption in the first place. However, if you can find a flaw in the encrption scheme to exploit, you might even be able to get by with less computation power.
>Even in high school, teach students how to describe problems as an optimization. Don't bother teaching them equation solving. Analytical solutions are sometimes needed, let that be a specialization for advanced undergraduates or even graduate school.
This would be a very, very bad idea. In order to use a tool like machine learning properly, you need a solid understanding of the problems you are trying to solve, so that you can frame them properly for the machine. Furthermore, there are many pragmatic real-world problems that require analytical solutions, much more than could be addressed by a small body of specialists. I think that basic programming concepts like iteration and recursion should be taught in secondary school (possibly even primary school), and I could definitely see adding basic concepts of optimization to that curriculum, but analytical thinking is so critical that taking it out would be an enormous mistake.
On computational power, we'll have to to agree to disagree. I believe self-driving cars and ng's helicopter are examples of why i think computing power is sufficient, there are many others (how kinect,google goggles,ibm watson etc etc. were built. There are also the "humie" awards in genetic programming which compete with traditional engineering). In engineering, finite element analysis has taken over. It is more human intensive than straight up machine learning but it's an example nevertheless of compute power displacing humans. I expect finite element analysis to be overtaken by machine learning too in many of it's applications as awareness/trust in machine learning gains mindshare.
Really, my basic point is, forget what you know, start afresh, and put faith in brute force searches. In natural language processing this happened 20 years ago and they made great progress, in computer vision this is happening right now, and these two fields are the toughest in my humble opinion. What they do, surely other fields can learn from.
There are too many hurdles to overcome. The psychological and liability one for actual plane manufacturers.
+ it just doesn't make sense economically. Pilots get paid fairly little now, and as automation progresses, the salaries will go down to ~40K(so ~80K for the 2). At that point, it's really not that much money to give passengers the extra sense of security.
As for the weight of the pilots themselves - judging by the cost of excess luggage, I'd say it's negligible for all but the smallest aircrafts.
Automation today is really good, but only until something goes wrong. Until automation becomes absolutely perfect, I will argue it's even starting to become dangerous - as pilots are thrown into emergencies without quite knowing how the plane got into them and based on what flawed assumption.
Even human pilots behave somewhat like automatons in some situations: in most situations they follow procedures, which could be described as "responding to inputs according to pre-defined instructions." However, they often encounter situations not covered by the procedures, in which case they must instead exercise their judgement.
The problem with judgement is that it is neither consistent nor reliable. Some humans have better judgement than others. Eventually we will have automation sophisticated enough to handle even to full complexity of aviation, at which point automation will yield safe results more consistently and reliably than human judgement.
A human being, on the other hand, has the capacity, when faced with unexpected or unfamiliar conditions, to exercise something we call "judgement" in an attempt to develop an appropriate response. I'm not saying that it's impossible for an automated system to have this capacity, I'm saying that no current automated systems have it, and that we're nowhere near to developing such a system any time soon.
I do believe they were planning to deploy it as a suggestion system though, which would suggest a course of action, and then leave it to the pilot to implement it or not. Then the judgment gets more murky; now the system is doing some of the judgment (evaluation of alternatives, etc.) that a human pilot would normally do, but leaving some of the judgment (accept the suggestion, modify it) still to the human.
edit: Here's a more recent paper than the one I'm thinking of, but must be the same project: http://ti.arc.nasa.gov/m/profile/de2smith/publications/IAAI0...
http://www.airliners.net/aviation-forums/tech_ops/print.main...
Plenty of other planes of lost a section of wing a wing and still landed. http://www.airliners.net/aviation-forums/general_aviation/re... Granted, all of these cases had a pilot, but in the F-15 the avionics actually discovered how to maintain level flight after the loss of the wing.
Automated flight control systems definitely do not exercise "judgement:" they have inputs, a transfer function (typically MIMO, these days), and outputs. It used to be that the transfer function was fixed, but more sophisticated systems (e.g. fighter jets) often have many different transfer functions and switch between them based upon various inputs. They don't "decide" or "discover" anything: for any given set of inputs, they will predictably produce a pre-determined set of outputs.
Aircraft stability and control systems are not programmed to care about, or even know about, the existance of the wings. The closest they get to this is that they will know the current states of the control surfaces on the wings. So it doesn't really make sense to say that the F-15 CAS was "not programmed for" the state of missing a wing (although it almost certainly was programmed to respond properly in the situation where it gets no feedback from some of the flight controls). As you say, it's designed to reach a specific goal (keep the plane level) and then maintain that state. If the system detects an uncommanded roll-rate, it will move the flight controls to stop that roll-rate. It doesn't know or care that the uncommanded roll-rate is the result of asymmetric lift due to an (almost completely) missing wing: it's just going to keep moving the flight control surfaces until that roll rate goes away. If the aicraft had been damaged in a slightly different way, it's possible that the CAS would have issued commands to the flight controls that would have departed the plane, but fortunately the handling characteristics of the aircraft remained close enough to normal that the control laws still produced good results.
Unfortunately, this behavior can result in mishaps when a flight control system gets erroneous inputs: when it believes that it is rolling when it is wings-level (or believies it is level when it is rolling). This was a major contributor to the Air France fligh 447 crash. In such situations, it takes judgement to realize that something is wrong and to figure out what to do about it.
However, while flight control systems have been responsible for plenty of crashes pilots have often mistaken level for non level flight and focused on faulty instruments rather than switching to working backups etc. An automated system can handle redundancy much more efficiently than people in such situations so while it's little value for a person to have ex:7 gyroscopes if they need to pick between them autopilots can gain from access to such information.
I was simply making the point that when you encounter a situation that isn't covered in the instructions, you need juman judgement to figure out the best way to proceed. I also explicitly stated that human judgement is far from flawless, and that if you can develop a sufficiently comprehensive automation program, you can get safer results than you would on average with human judgement.
PS: IMO, what separates people from machine learning systems is treating everything as training data, a much larger training set, a lot more processing power, and a tendency to explore novel situations. The trade off is efficiency and reaction times. Still, when you get into thrust vectoring, super sonic flight, high g turns, rapidly changing weight/drag/thrust at the same time trading consistency for improved handling of novel situations is probably worth it so I expect the air force uses systems that are fare more adaptable than the civilian world.
Where are you getting this definition of "automation" from?
As far as I'm concerned, an automatic system is simply one that require little or no direct human control.
There are many, many aviation disasters that were avoided because the pilot got creative.
Approved aviation systems are far from bleeding edge; at best the technology in use is five years old, and the 5-year old technology that's in use is only so because some company(ies) invested the considerable resources to jump through all of the hoops required by the FAA to get approval for commercial use.
Furthermore, military aviation is an incredibly poor analogue to commercial (and general) aviation because it lacks the onerous regulatory oversight brought to bear by the FAA. Military aviation is the bleeding edge and the advancements made there trickle backwards to CA/GA at a very slow pace due to the incredible cost of getting certificated by the FAA as well as the fact that the FAA (correctly) places safety and reliability at the top of their priority list.