Unfortunately I had to come back to have a look at this thread. I shouldn't have.

Quoting Max Q (Reply 176): Nobody wanting Pilotless Aircraft has everything to do with whether this concept will be accepted ! but why stop at Pilots, why not do away with Air Traffic Controllers, how would you like driverless Police cars / Robocops as a next step ?

I'll say it again, and then probably won't bother again... the original discussion was about the CONCEPTUAL FUTURE. I completely agree with 95% of the points made by yourself and Pihero, when applied to TODAY - but you're not looking nearly far enough ahead. This has been said 100 times already, but 101 won't hurt - you are arguing the impossibility of a situation that could occur 100 years from now by discussing TODAY'S technology. The argument is simply not relevant.

We could be having a similar discussion about when normal RPT services will be flown at hypersonic cruise speeds - and people would be calling it an impossibility, and using the current limitations of jet technology and aircraft design to prove their point - and the arguments would be absolutely irrelevant. How can anybody in their right mind claim that there is NO WAY a hypersonic cruise could be standard practice 100 years from now? Or 150 if you like? Pick a number. It's quite possible, depending on the developments over decades to come.

The EXACT same logic applies to this discussion. There are quite a few people here who are ridiculously sure of how technology, software and artificial intelligence will develop over the next century or so - according to them, it won't really progress much at all. Weird, considering the exponential advances in technology over the past 50 years. I wonder how many of those advances were labelled "impossible" when they were first tabled.

And please don't anybody lump me into some "pro-computer" basket. I'm a pilot first, software engineer last, and I sincerely hope we get nowhere near pilotless airliners in my lifetime. But my background in computer science and software engineering has allowed me to conceptualise past next week, and actually think about clear possibilities 20, 50, 100, 150+ years from now ("science fiction" to a couple of you) - and all the concepts here are entirely possible, if we don't put a limit on the development time.

Ok, enough from me.

Quoting Rheinwaldner (Reply 200): When Will Computers Take Over Pilots? Never!

You've just proved my above point. You've quoted a heap of limitations of current technology - probably all very correct points - and then used it to prove a case 100+ years from now as impossible. Even more so - you've used recent incidents, and then stated what would happen if we just removed the pilots and let the aircraft go on ahead by itself. This is all irrelevant to the original discussion.

So, as a software developer, do you think we've reached the pinnacles of artificial intelligence and software engineering? Because in my opinion, that'd be a pretty crazy statement.

Quoting Zappbrannigan (Reply 201): Even more so - you've used recent incidents, and then stated what would happen if we just removed the pilots and let the aircraft go on ahead by itself. This is all irrelevant to the original discussion.

No, I listed cases where no artificial system could have resolved the situation.

As an engineer I can overlook how a system would look like that would cope with all this conditions automatically. The reason why it does not work is not because so far unheard technology will provide a solution in 100 years. Any of these problems can be solved with today technology. But system complexity would rocket up. Not on the computer side (here the power is tremendous and probably sufficient), but on the actor, sensor, network and redundancy side. Some of pilots ability are indispensable at all (e.g. retracting a gear by manual crank).

The core problem are system failures. By nature systems can never throne in all circumstances over system failures. Systems that should not fail fail all the time. Systems preventing other systems to fail or offer workarounds for other systems come only at extremely high cost in an environment like an airplane. Today systems are much less complex than what would be required to steer planes automatically. And even though they are made to be bullet-proof they fail by themselves or because of outside interruption all the time. No technology enhancement will ever deliver systems with sufficient MTBF to fully control aircrafts. The magnitude of involved technologies, parameters and systems will ever need man to supervise incidents.

Even the way from the gate to the runway is full of deadly traps if there is not a man in the cockpit overlooking the situation.

Your logic is sound - but your arguments are still based somewhat on today's standards. Yes, system complexity would be ASTOUNDING. This absolutely cannot be argued. But is this not conceptually possible? That what we think of as astounding complexity will be household junk in 100 years? (For the record, this 100 years I keep saying is just a random number to illustrate I'm not talking about next year). I don't have a problem with your logic, you obviously know what you're talking about - but I do not agree with your use of the word "never". I can quite possibly believe that computer systems will become so reliable as to be, for all intents and purposes, fail-proof. But to see that ever happening, one needs to stop looking at today's limitations as a near-limit.

I simply cannot agree with the logic that "well we couldn't achieve that level of reliability and system complexity in the next 10 years - so it's technologically impossible, ever.".

Quoting Zappbrannigan (Reply 203): I do not agree with your use of the word "never".

Indeed. Historically, when people have said "this can never happen", they have often been proven wrong.

My father was once told by a car engine designer that five cylinder engines were "a technical impossibility".

I bet that 100 years ago if you told people about the A380 most would have said that it was impossible.

Rhinewaldner you bring up some very good points, it is good to see another rational voice on this subject.

Just because something is conceptually possible does not make it a good idea.

Certainly we are talking about todays level of technology but that is irrelevant, no amount of technology will replace the flexibility of the human mind.

Rheinwaldner's examples should give you pause, how would a Pilotless aircraft react to a potential mid-air with a non transponder equipped aircraft, the landing with nosewheel cocked ?

As another poster mentioned, how would it deal with a routine weather deviation, weather radar returns take skill to interpret.


You know Scientists have been attempting to come up with a high tech device to replace a Dog's nose for years with no luck, you literally cannot improve on Nature in some cases !


Unexpected situations will always occur, failures thought of as 'inconceivable' will happen.

You cannot 'engineer' in: Judgement, Experience, Training, Intuition and last but certainly not least, Handling skills.

Quoting Max Q (Reply 205): Just because something is conceptually possible does not make it a good idea.

Mate, could not agree with you more. I am discussing this from a "future engineering and systems capability" angle - not from a "what's best for aviation" angle.

Personally, I'd love to see technology continue to advance in leaps and bounds - but always with two crew up front, even if a normal operation becomes completely hands-off from push to parking, to keep a human face on aviation at all levels, and to allow the world's greatest job to continue.

Quoting Max Q (Reply 198): Manual reversion is not 'my definition' it is Boeings

You defined it two ways...which one is Boeing's definition?

Quoting Max Q (Reply 198): .I have flown manual reversion drills, have you ?

Not on a large airliner, no. Which has nothing to do with whether your definition of "manual reversion" is correct or whether the 777 (or anything else) has it.

Quoting Max Q (Reply 198): If you really understood ADS-B and CPDLC you would know it requires humans on both ends of the data link, as an example how could the respective FIR control authorise a climb or descent en route without human response

Why would you need a human response? Today's system just presents the messages to the pilot, but since it's trivial to send standard format messages, there's no particular reason the computer can't directly load them. If you just automate the "accept" button you'd have the whole flight plan loaded without any human involvement in the flight deck.

Quoting Max Q (Reply 198): Even if your computers don't make mistakes it doesn't matter, if the situation develops it has no program for, if it loses it's power supply etc etc a perfect computer becomes an expensive paperweight.

Very true.

Quoting Max Q (Reply 198): Where you seem to be fundamentally unable to get the point is your blind faith that all error can be 'engineered out'

Quoting Max Q (Reply 198): Even if you design a 'clean sheet' autonomous aircraft errors will be made

Quoting Max Q (Reply 198): You are truly Naive if you believe you can 'eliminate the potential for Pilot or computer error'

Quoting Max Q (Reply 198): Unless you have seen this in real life I suppose I can see how you can delude yourself into believing computers and automation hold all the answers.

I'm very puzzled by your apparent refusal to parse these statements:

Quoting Tdscanuck (Reply 196): NOBODY (except you) IS CLAIMING COMPUTERS ARE OR WILL BE PERFECT. EVERYBODY (including you) EXPECTS COMPUTERS TO CONTINUE TO HAVE FAILURES.

What about this is not clear to you?

Quoting Max Q (Reply 198): You argue that recent, quite hazardous automation failures happened because humans were designed to be 'in the loop' to take over if necessary.

No, I didn't argue that at all. Those failures happened, typically, because of design or component issues. The reason we *accept* those types of failures from those systems is because of the human in the loop.

Quoting Max Q (Reply 198): and, unlike the Mas777 and Qantas A330 there will be no one there to turn off the stupid computer and hand fly using the invaluable skills of human Pilots, another point you repeatedly choose to address

I've addressed it many many times. You choose not to accept my arguments, which is certainly your right, but let's dispense with the idea that it hasn't been addressed.

Quoting Max Q (Reply 198): So, programmers would find it 'pretty trivial' to write programs for an aircraft a human could never execute.

Sure...you show me a human pilot that can pull a sustained 15 g turn.

Quoting Max Q (Reply 198): That's probably true, I don't know any Professional pilots that would make such a violent control input in cruise that many of their Passengers are seriously injured.

You mean besides the A300 over NYC, or the Braniff 707 in Seattle?

Quoting Max Q (Reply 198): EGPWS was not in existence when AA crashed in Cali, I have used it for real in mountainous terrain, I have also seen it fail, as can any system. A proficient human Pilot, however would not rely only on that ,but would be constantly verifying his position not only with the FMS but also on his High and Low level charts.

If a proficient human pilot would be constantly doing these things, how on earth did the AA plane crash? Are you suggesting AA runs un-proficient pilots?

Quoting Max Q (Reply 198): 'Total navigational failure is unheard of' I wish we had known this when we did, indeed lose all lateral and vertical position in our 757 about 8 years ago over the Atlantic. It took us over an hour using many unorthodox 'out of the box' solutions to restore just the Lnav (never did get the Vnav back)

I meant navigation as in "where is the airplane", not lnav/vnav as in "the autopilot modes". A/P modes drop out all the time and it doesn't surprise anybody. Simultaneous loss of GPS, VOR, DME, IRU, and dead reckoning position data from all systems is extremely rare (I didn't say "unheard of," I said "essentially unheard of").

Quoting Max Q (Reply 198): We had another 757 lose all electrical power on a flight out of Anchorage including, eventually standby power, the crew made a safe landing in Ketchikan, they were actually physically holding the APU switch in the start position to 'force' power onto the essential buses .This, once again shows your lack of real life experience and misguided trust in computers.

How does it show lack of real life experience or misguided trust in computers? Nobody is claiming these types of things don't happen on current airliners. As a non-FBW airliner, the 757 has a considerably laxer electrical power requirement anyway.

Quoting Max Q (Reply 198): But you go on to show even more ignorace.. 'a computer can certainly execute any particular flight path more accurately and consistently than any human' have you ever watched an Autoland in limiting wind conditions ?

Yes, I have. You are deliberately ignoring the difference between "can" and "does." Using today's technology, an autopilot can access the complete range of flight control inputs that a human pilot can...that means they are capable of employing the same control forces. Their positional awareness is significantly better, especially with regards to transient acceleration. And their reaction time is much much faster.

Quoting Max Q (Reply 198): The autoflight system is simply incapable of performing safely in these conditions, you are basically just guessing with such a statement, my fellow Pilots and I have seen this in real life.

Just because my opinion differs from yours doesn't make it a guess. As I suggested a few times before, you might be well served to do some research on where the information comes from. Current autoflight systems perform just fine within their design limits and (surprise surprise) are not safe outside their design limits. This isn't exactly shocking to anyone.

Quoting Max Q (Reply 198): Why not enlighten me now on the reasons for wind limits on Autoland then ..

Because current autopilot design gives the autopilot less control authority than the human has. Note: we're talking design now, not technology. Current design autopilots are incapable of making the same control inputs that the human pilot can. That means the human pilot can operate in a considerably wider operating envelope. However, that's a conscious design decision on the part of the autoflight folks, not a technical or physical limitation.

Quoting Max Q (Reply 198): And, on the contrary, I, along with my fellow 'real world Pilots' are most certainly at an advantage with respect to automation and control as we have been using it for real for decades

I suspect you've been using a PC for decades too, but that doesn't mean you're at any advantage with respect to microprocessor design.

Make no mistake, I'm absolutely willing to defer to pilots on issues that pilots have the most knowledge of...like how current airliners behave, what situations can and can't reasonably be handled by a human, etc. I don't have any argument with your experiences regarding automation failures or performance limitations.

However, you need to realize that your experience base is entirely centered on current technology airliners. These are not designed to ever operate pilotless and are not the airliners that the OP asked about, therefore the relevant experience for what's possible is not what we have now, but what we can have in the future.

Quoting Max Q (Reply 198): Just one last question, would you put your family on a Pilotless aircraft ?

In as much as I can speculate about a product that doesn't exist, yes I would. I put them on manned airliners all the time knowing full well that those airliners (both the plane and the crew) can fail. The fact that an automated airplane can also fail isn't a meaningful metric of anything.

Quoting Zappbrannigan (Reply 199): Unfortunately I had to come back to have a look at this thread. I shouldn't have.

Get out while you can!

Quoting Rheinwaldner (Reply 200): True, human error has caused many accidents. On the other hand uncounted (and much more) situations have been ironed out by human cognition.

This is more like it...if it's much more, how do we know? I'm not disagreeing with you, this is just a very difficult data point to get any accuracy on so if you've got some numbers to put to it let's get them out in the open.

Don't forget we also need to include human errors that have been ironed out by the automation...my gut feel says this is a smaller number, but it's even harder to get this data than to get computer failures saved by the humans.

Quoting Rheinwaldner (Reply 200): The problem with automated planes is that you can't foresee any condition that needs to be addressed. If the computer has the built in capability to solve anything that has happened so far (IMO already unrealistic) it still does not know the condition that may arise tomorrow for the first time. I assume such situations arise daily and we never get to know ... because pilots apply some unorthodox combination of methods.

We're back to FMEA here...I don't think anyone with even a trivial grasp of fault management would expect to design a pilot less airliner based on exhaustive fault trees alone. Besides being a lousy design philosophy, it's probably an untractable design problem, as you noted.

Quoting Rheinwaldner (Reply 200): E.g the A320 with the front gear steered to the side. Just estimate the system that would have to be in place to cope with that situation.

It's actually relatively trivial...it's just a SIMO controller that takes the desired yaw rate as input and drives the rudder, thrust, spoilers, nose wheel steering, and brakes as outputs. This type of control problem is extremely well studied and not all that difficult...there are much more challenging issues facing a pilotless airliner.

Quoting Rheinwaldner (Reply 200): Other examples: How can an automated plane solve something like this? - ATC reports (as personally witnessed): "Can you see the (obviously transponderless) towplane crossing your flight path?"

The specifics obviously aren't ironed out, but I have to assume that a pilotless airliner needs some kind of local situational awareness system. Shortwave radar is a fairly obvious choice, but you could also use LIDAR at the right wavelength.

Quoting Rheinwaldner (Reply 200): - Whenever unusual system failures happen, e.g mechanical damage

Are you thinking of aero damage or something different? The former has already been pretty thoroughly addressed on fighters and some NASA research.

Quoting Rheinwaldner (Reply 200): - "Soft decisioning" -> e.g. selection of alternates in delicate cases

This, indeed, one of the most challenging cases. It may be one where the computer can never catch up.

Quoting Rheinwaldner (Reply 200): - Whenever pilots have to react on visual or aural evidence. Even unusual smells cause frequent returns to the airport. How to teach a computer to smell? Where to place the sensors? Everywhere?

This isn't much different than smoke detection requirements on freighters...other than adding more comprehensive sensors, I don't see why it's a major hurdle.

Quoting Rheinwaldner (Reply 200): - Conflicting raw input data that could be resolved by a simple look out of the window.

This is another reason why you need local situational awareness.

Quoting Rheinwaldner (Reply 200): - Situations that are solved in a negotiating way with interactions between pilots, ATC and maybe other aircrafts.

At least in any foreseable generation of pilotless aircraft, I think there's broad assumption that you need to revert to UAV-style remote control under some situations. This seems like one of those.

Quoting Rheinwaldner (Reply 200): Lost aircrafts if no pilot would have been on board: - Turkish A310 approaching Lagos and having to divert in the night with a complete loss of navigation. Landed on an unknown airport that accidentally left on the runway lights. -> Would have been biggest crash this year. - I'm not sure whether the diverse troubled Qantas flights would have ended so well without pilots on board. - There was a LX saab 2000 landing finally on a closed air base after running out of fuel

The Qantas flights span a pretty wide range of failure types, so that one is probably best dealt with on a case-by-case basis. For the Turkish and LX cases, I would have to assume that any pilotless airliner would include suitable landing areas, not just active airports, in the nav database.

Quoting Rheinwaldner (Reply 202): Systems that should not fail fail all the time.

"All the time" is very subjective...when was the last time we know of a full FBW or dual engine failure that wasn't common mode? As noted in some prior posts, there is a very wide range of allowable reliability on aircraft systems, depending on what you're talking about.

Tom.

Quoting Max Q (Reply 205): You know Scientists have been attempting to come up with a high tech device to replace a Dog's nose for years with no luck, you literally cannot improve on Nature in some cases !

Well, it can perhaps be done. We just need to try harder.  

If you worship machines, they will most certainly be your master !

[Edited 2008-10-18 01:38:09]

Quoting Max Q (Reply 209): If you worship machines, they will most certainly be your master !

Now you're just being preachy.   I have notice that those who work closely with computers tend to find them much less "magical" than those who do not.

Quoting Max Q (Reply 205): You know Scientists have been attempting to come up with a high tech device to replace a Dog's nose for years with no luck, you literally cannot improve on Nature in some cases !

...and if we were discussing the possibilities of a fully automated commercial canine, you might get different arguments.

Quoting Max Q (Reply 205): how would it deal with a routine weather deviation, weather radar returns take skill to interpret.

Nothing that a computer couldn't interpret ... after all, the computer created them, remember  

Quoting Max Q (Reply 205): how would a Pilotless aircraft react to a potential mid-air with a non transponder equipped aircraft,

The same way you do ... you 'see' and avoid. Cameras, FLIR, Primary RADAR (or similar) could all be used as inputs for the systems to perform avoidance maneuvers. In fact, even antiquated primary radar systems can do a better job of identifying a conflicting target than a human's eyes ... that may or may not be looking at the right part of the sky.

Jimbo

There's a pair of very interesting articles that relate to this topic in this week's AW&ST.

One is about the new benefits vs. new risks of technology and automation. The FAA aviation safety admin states outright that "Technology has been the greatest contributor to the reduction of accidents, bar none." They go on from there to discuss how increasing automation also poses unexpected hazards, as we've been discussing, including the difficulty of exhaustively testing very complex systems. On the whole, it's pretty neutral on the thread topic.

The second article is much more concrete and goes into some of the latest advances in automated control. A DARPA program at Rockwell Collins' now has a system that successfully managed the following test:
-Go to fully automated flight
-Blow off 60% of the wing area

The system returned to controlled flight in 1 second and recovered baseline performance in about 1 minute (i.e. it has performance equal to the original *while missing 60% of the wings*), then was able to stick a crosswind autoland with 20 centimeter tracking error. The most interesting part is that they system was never programmed with how to control the airplane with 60% of the wing missing (or any %). It just has two control models...one that simulates what it wants and one that watches what the airplane is doing...the two talk to each other to adjust the control gains until the performance matches. Such a system can control any controllable aircraft without prior specific failure programming, after getting to play with it on the order of a minute. That's the type of system that I have to assume would be part of any notional pilotless airliner. That is also, incidentally, considerably better performance than any human pilot on record has been able to achieve under similar types of control loss.

The experts (and these are experts) acknowledge that the final outcome is unclear. Collins' stated goal is to make UAV's as reliable as piloted aircraft so that they can operate in civilian airspace without restriction. So at least somebody (and a pretty big somebody) is working towards the goal.

The above is just a summary and, obviously, subject partly to my paraphrasing. I recommend that anyone actually interested in the current state of the art read the both full articles (pages 48-53 in this week's issue).

Tom.

Quoting Zappbrannigan (Reply 203): but I do not agree with your use of the word "never".

True, in fact it is my saying that there are words you should "never" use ... and one of them is "never"!

Quoting Tdscanuck (Reply 207): It's actually relatively trivial...it's just a SIMO controller that takes the desired yaw rate as input and drives the rudder, thrust, spoilers, nose wheel steering, and brakes as outputs. This type of control problem is extremely well studied and not all that difficult...there are much more challenging issues facing a pilotless airliner.

How about the optimum configuration? Could be all gears in, but if the danger is extreme break away to one side (e.g. because the front wheel is stuck at an 45° steering angle) an idea could be to extend only the main gear on the other side. This would probably offer the best chances of sliding forward. In such cases the crew decides about best chances. There is sometimes no hard true and false. An area where computers are very limited (the limits are exactly set by the imagination of the programmers).

Quoting Tdscanuck (Reply 207): This isn't much different than smoke detection requirements on freighters...other than adding more comprehensive sensors, I don't see why it's a major hurdle.

It's not only smoke. That system should be able to discriminate burnt coffee or other harmless smells from those coming from critical sources. Very difficult to obtain valid conclusions by automated systems. That system must have knowledge of the duties executed by F/A's in the last 15 minutes to classify the severity of "smokes". Realistic?

Quoting Tdscanuck (Reply 207): For the Turkish and LX cases, I would have to assume that any pilotless airliner would include suitable landing areas, not just active airports, in the nav database.

For LX: The airport was not just inoperative in was actually broken. A wall of mud crossed the runway. It was enough suitable for the passengers to survive. It would not have been in any database.

For TK: Problem was loss of any navigation system. What use has any database in this case?

I come to the conclusion that most of today emergency landings require the presence of pilots onboard. In most cases the reason for emergency landings are failed systems. For fully automated planes therefore a paramount improvement in system MTBF must be achieved. Otherwise nearly any of todays emergency landings would be crashes.

Quoting Tdscanuck (Reply 212): The FAA aviation safety admin states outright that "Technology has been the greatest contributor to the reduction of accidents, bar none." They go on from there to discuss how increasing automation also poses unexpected hazards, as we've been discussing, including the difficulty of exhaustively testing very complex systems. On the whole, it's pretty neutral on the thread topic.

Now that is a different and very promising target. IMO safety has been greatly improved and will further in the future by validation-, assessment- and prioritization-systems. I think the recent Madrid accident should have been prevented by a configuration checking system. It is time that no planes will ever be moved again in the wrong configuration. That is a very good candidate to implement an automated system that prevents such things once and for all.

Generally for me it is not the question "what" could ever be achieved. As I said the technologies to "do" most are already available. Beside severe things that need extremely costly installations to just provide a lousy substitution of human capability on board there are many issues with the cost-benefit balance. The reason why absolute words like "never" have some legitimation in this discussion is because the technologies to realize the pilotless plane really do exist. In the future trade off studies will therefore very likely not come to a fundamentally other conclusion.

You must be aware that not in every area technology will ever improve. On the opposite any problem to be solved by technology once will be "solved". After that time progress will stop in that particular area. Many technical solutions are "the" answer for a need (sometimes since a long time, some even for centuries) and will not be further developed. Also in aviation many breakthroughs are settled. The majority of systems to substituted a pilot are available. And still on the cost and effort side there is no benefit to implement all these sensors, networks and computer power in an airplane. And there is no fundamental breakthrough in sight that gives us the smallest hint of changing that premise. Ask electrical engineers to improve the availability of their components by multitudes. You will earn shaking of the head. And yet without such devices no pilotless plane will ever fly with passengers.

In my mind I can easily layout systems from sensor to actor. I am aware of boundary conditions, racing conditions, availability considerations, scalability, cost of stuff and maintenance. As electrical engineer now working as software developer with many fancy ideas (really believe me: I am often suspected of being extremely unrealistic! Even on this board I went out of a limb with some threads I started) I believe to have a sound understanding of good and bad automation candidates. Replacing pilots is a very bad one!

Quoting Rheinwaldner (Reply 213): That system must have knowledge of the duties executed by F/A's in the last 15 minutes to classify the severity of "smokes". Realistic?

Just automate the F/As. Duh.  

Another very well reasoned anaysis Rheinwaldner, but for some you may be clouding the issue with facts !


It is certainly true that technology has brought great advances in safety in Aviation and there are many more promising developments on the horizon.


But that technology is no panacea, as most of us realise there are endless permutations of different problems in flight and always will be, problems requiring the flexibility of a human response.




Whether you like it or not there is simply no replacing our skilled Pilots.

Quoting Starlionblue (Reply 214): Quoting Rheinwaldner (Reply 213): That system must have knowledge of the duties executed by F/A's in the last 15 minutes to classify the severity of "smokes". Realistic? Just automate the F/As. Duh

Why would the system need to have the knowledge of the F/A's duties?

The F/A will still exist in this scenario, otherwise who is burning the coffee? ... and that is the person making the distinction between burnt coffee or burning aircraft...

Regardless, it's quite trivial for a sensor to distinguish different 'kinds' of smoke .... it doesn't need a canine nose.


Jimbo

Quoting Bond007 (Reply 216): The F/A will still exist in this scenario, otherwise who is burning the coffee? ... and that is the person making the distinction between burnt coffee or burning aircraft...

That would work!? To continue your line of reasoning: The consequences are buttons in the galley:
"I have burnt eggs", "I have burnt coffee", "I have burnt fish" and the cabin staff now can overrule the fully automated safety system, what a comforting thought!

BTW you are aware that there had to be a button for cabin staff to be used to approach the closest airport in case of medical emergencies? Of course appropriate selection based on the type of the disease, and X other criterias should possible too.

Guys, this is just an example of the difficulties to make everything automated. This is not the end, it's just the beginning. It just shows you get nowhere if you start to automate "everything".

The proposals so far don't solve the problem of fire on board. What came up so far either would causes crashed planes or needless returns to the closest airport because of false alarms.

Quoting Tdscanuck (Reply 212): The second article is much more concrete and goes into some of the latest advances in automated control. A DARPA program at Rockwell Collins' now has a system that successfully managed the following test: -Go to fully automated flight -Blow off 60% of the wing area

Tom,
I read the article and was duly impressed. But I have two comments :
1/- That event occurred sixty five years after very little experienced youngsters managed to bring back home a few "un-flyable" bombers which have suffered horrific damage.
2/- An F-18, like all modern *combat* aircraft are built both for manoeuvrability and for damage tolerance, to the point that it is quite difficult to know where the roll / yaw /pitch controls start and where they end. On any airliner you can think of, you'd have dedicated controls... and unless you design future generations of air transport airplanes with that goal, (and why not, apart from the weight and cost considerations, which are at least very important ?) the demonstration is not valid.
Proof is this video called "F/A-18 Hornet vs A-4 Skyhawk" which you'll find
at this site (the site is quite gorgeous, btw, that's why I provide the link) in which the loss of about 1/5th of the wing sends the poor A-4 in an uncontrollable roll and spin, and I would compare the control surfaces of a Skyhawk to our airliners'.

Now, for the proponents of "all-automation-forward", this little link about the X-43 which is as far as I understand sponsored by both NASA and DARPA :
Cause of X 43 loss,
from which I extract these gems :
"...the failure of the NASA X-43A hypersonic aircraft in June 2001 was the result of inaccuracies in computer and wind-tunnel tests that were based on insufficient design information about the vehicle itself...
According to the MIB documents, the X-43A Hyper X launch vehicle "failed because the vehicle control system design was deficient for the trajectory flown due to inaccurate analytical models (Pegasus heritage and Hyper X launch vehicle specific), which overestimated the system margins...
The X-43A mishap board points to a series of "modeling inaccuracies" used in ground tests, as well as "misinterpretationn" of wind tunnel data due to insufficient data.
... The board found that the flight mishap could only be reproduced when all of the modeling inaccuracies are taken into account in post-failure work dedicated to unraveling the cause of the X-43A mission loss....

Quoting Tdscanuck (Reply 207): Quoting Rheinwaldner (Reply 200): E.g the A320 with the front gear steered to the side. Just estimate the system that would have to be in place to cope with that situation. It's actually relatively trivial...it's just a SIMO controller that takes the desired yaw rate as input and drives the rudder, thrust, spoilers, nose wheel steering, and brakes as outputs. This type of control problem is extremely well studied and not all that difficult...there are much more challenging issues facing a pilotless airliner.

I think that you underestimate the problem :
1/-Nose wheel steering is lost
2/-Rudder could be effective but will lose that effectiveness under some 60 knots.
3/- If you use the brake and the spoilers ( btw these are still symmetrical as ground spoilers) for directional control, you lose them effectively as deceleration devices...and as reverse thrust won't be that efficient, would you consider using forward thrust to counter the yawing moment of the nosewheel ?......
So let's see : Loss of half the braking power + one engine in high forward thrust + no spoilers =....runway overrun anyone ?
Please think again, because if I were in that situation, I would call you engineering folks to advise me on all the systems idiosyncrasies...and your initial thought is the very one I would have rejected in the first place.

What I also see is that WE COULD design a fully automated airliner... but "could" is not "can safely" and so far to all the arguments Max, Rheinwaldner and others have said, the answer has been mainly " more sensors, more info" which , in our to-days modern technology leads to a massive complexity of the system,a lot of weight increase and some major hurdles to overcame : for instance, I'd really like to see hundreds of aircraft sharing the same airspace, each one radiating radar pulses, LIDAR waves plus two dimensional cameras looking around all sectors around the aircraft....let's be serious, our finance managers would have a fit at the cost, plus the one of hiring countless of new engineers and technicians ( which is still in my mind the whole subject of this thread)...
In fact I share ZappBrannigan' sentiment :

Quoting Zappbrannigan (Reply 206): Personally, I'd love to see technology continue to advance in leaps and bounds - but always with two crew up front, even if a normal operation becomes completely hands-off from push to parking, to keep a human face on aviation at all levels,..

(My stress on *normal*)

Quoting Rheinwaldner (Reply 217): That would work!? To continue your line of reasoning: The consequences are buttons in the galley: "I have burnt eggs", "I have burnt coffee", "I have burnt fish" and the cabin staff now can overrule the fully automated safety system, what a comforting thought!

That's exactly how you described the 'human' scenario earlier ... there is a smell in the the galley, and the F/A made a decision on whether it was burnt coffee, or burnt wiring insulation, and forwarded that information to the cockpit ... whether there is a human or a computer behind the cockpit door is totally irrelevant.

Quoting Rheinwaldner (Reply 217): The proposals so far don't solve the problem of fire on board. What came up so far either would causes crashed planes or needless returns to the closest airport because of false alarms.

They didn't? In fact, all of the proposals solved it, just not to your personal satisfaction. There are numerous types of smoke/gas sensors available that are far more capable of a decision on whether the smell is 'serious' enough to warrant an emergency landing, than an F/A's nose.
Really, this is a minor issue in the whole big scheme of things!

Quoting Pihero (Reply 218): let's be serious, our finance managers would have a fit at the cost, plus the one of hiring countless of new engineers and technicians ( which is still in my mind the whole subject of this thread)...

Well, part of the discussion was on whether the systems and technology was capable of controlling a fully automated aircraft. We have all agreed that there are many other factors that may hinder the implementation .... but not that it isn't technically feasible.

Jimbo

Quoting Max Q (Reply 168): We are all 'anonymous' on this forum, all the more reason why you should have no hesitation in being able to provide some evidence of the reality of your 'thousands of hours experience'

He could simply lie. We have no guarantee that you are what you say you are. The fact that he does not answer to your ad-hominem baiting says as much about him as the baiting itself says about you.

As for pilotless aircraft, it could be done today, with existing technology and at a level of reliability consistent with certification standards for aircraft systems (25.1309 in particular). The cost, public acceptance and regulatory acceptance make this it an unattractive investment. Which is why no one is risking their money on it.

Given these factors, aircraft will instead evolve with computer systems taking over more and more of the pilots' functions. We already have the AP engaged 95% of the time, already have auto-landing. In time there will be auto-TO, auto-taxi and others.

The pilot will stay in the cockpit but will become more a systems operator and decision maker than a pilot. At some point aircraft will have an autonomous "get me home" mode and, when all these evolutionary features are mature, we may see single pilot operations.

Whether having no pilot at all is an option after that is up for debate - it depends a lot on how well a remote operator can deal with unexpected events. No computer can emulate human decision making under unforeseen conditions.

Quoting Mrocktor (Reply 220): The pilot will stay in the cockpit but will become more a systems operator and decision maker than a pilot. At some point aircraft will have an autonomous "get me home" mode and, when all these evolutionary features are mature, we may see single pilot operations.

It already exists in GA aircraft, there was an article I pointed out earlier in which the eCFI as it was called noticed the aircraft was drifting off course. It aurally alerted the pilot then activated the stick shaker, but there were no control forces on the stick so it assumed that the pilot was unconcious. It then selected a suitable alterntve, contacted the tower, and made an emergency landing to awaiting rescue services. And, this entire system was created by a retired electrical engineer.

Quoting Max Q (Reply 215): Another very well reasoned anaysis Rheinwaldner, but for some you may be clouding the issue with facts !

Max Q, look up the definition of facts. A fact is a true statement, regardless of whether it agrees with your opinion.

Quoting Rheinwaldner (Reply 217): That would work!? To continue your line of reasoning: The consequences are buttons in the galley: "I have burnt eggs", "I have burnt coffee", "I have burnt fish" and the cabin staff now can overrule the fully automated safety system, what a comforting thought!

We trust them to get us out of an emergency, perform first aid, why can't we have them prevent emergencies?

Quoting Rheinwaldner (Reply 217): BTW you are aware that there had to be a button for cabin staff to be used to approach the closest airport in case of medical emergencies? Of course appropriate selection based on the type of the disease, and X other criterias should possible too.

Its not the pilots' job to diagnose diseases, all they have to do is take the opinion of medical professionals and find a suitable diversion if neccessary. The medical professionals wouldn't be replaced, the FA could talk to a ground-based doctor, and then tell the plane to find the best diversion within x-minutes.

Quoting Pihero (Reply 218): Proof is this video called "F/A-18 Hornet vs A-4 Skyhawk" which you'll find at this site (the site is quite gorgeous, btw, that's why I provide the link) in which the loss of about 1/5th of the wing sends the poor A-4 in an uncontrollable roll and spin, and I would compare the control surfaces of a Skyhawk to our airliners'.

I'm not sure what you meant by that example, other than that a human-controlled A-4 crashed when it lost most of its wing. His example showed that even unpredicted damage can still be controlled by computers thats not as simple as "if this, then that".

Quoting Pihero (Reply 218): "...the failure of the NASA X-43A hypersonic aircraft in June 2001 was the result of inaccuracies in computer and wind-tunnel tests that were based on insufficient design information about the vehicle itself...

Scramjets are still not fully understood and supersonic combustion is still extremely difficult. You can't take loss of an unmanned, highly experimental aircraft as a point against automation. The engineers made mistakes, and have since learned from them, and I beleive the X-43 has made successful hypersonic flights since then.

Quoting WPIAeroGuy (Reply 221): I'm not sure what you meant by that example, other than that a human-controlled A-4 crashed when it lost most of its wing. His example showed that even unpredicted damage can still be controlled by computers thats not as simple as "if this, then that".

Oh ! Please ! Show me how that airplane minus wing, ailerons and spoilers could have been controlled !
I wanted to show that what was achieved by Rockwell-Collins required a totally different flight control architecture than is usal to all airliners I know.

Quoting WPIAeroGuy (Reply 221): Scramjets are still not fully understood and supersonic combustion is still extremely difficult. You can't take loss of an unmanned, highly experimental aircraft as a point against automation.

1/- The accident had nothing to do with the scramjet but with flight controls.
2/- That example showed -and that was the point- that the design was faulty not to a lack of data but to a lack of having considered all the factors, and (quote)an overestimation of the margins(unquote). It means that humans, be they pilots and engineers are susceptible to errors (in this case by omission).

Quoting WPIAeroGuy (Reply 221): You can't take loss of an unmanned, highly experimental aircraft as a point against automation.

I dare say Tom's example fits very well to that description, too. Another case for the goose and the gander, perhaps ?
And I say it again : I am not against automation, I am against fully autonomous systems. The main reasons are :
1/- Automation so far is only good for procedural problems and certainly not able to cope with what Tom and Rheinwaldner call "soft decisions", which are the major part of flying an airplane.
In the example of the nose wheel stuck at a 45° angle from straight is also telling : The safest solution is a belly landing, as all others are linked to a degradation of the airplane performance -and evacuation-.
Problem is : that solution requires a certain amount of lateral thinking and considerations outside the technical side, which engineers don't do generally in this kind of situation.
Look for instance :
MELed item : APU inop...aircraft dispatched to a place in North Africa where the temperature is 40° C +, prevision of 185 passengers on the return leg. I called both dispatch and engineering : I'm not taking that airplane there as there won't be any ground conditioning cart available and that temp is going to cause havoc on my passengers, my crew, my electronics, my performances...They came quickly to my conclusions and the flight was delayed for two hours for an airplane swap.

2/- Replacing pilot mistakes by engineering design errors or oversights ( the result is the same : a faulty system) is not exactly what I would call progress.

3/-Contrarily to what some people are saying, we have reached a plateau on flight deck automation and I can't really think of any change since the 744 which flew nearly thirty years ago...Better communications, perhaps, better flight controls ( à la Airbus A380) and what else ? The EGPWS in my opinion poses more problems due to its dependance on accurate mapping and positioning, than the GPWS, TCAS has brought so many false alarms in terminal areas that it runs the risk of being disregarded by some crews (which is not really the finality of the project is it ?)...and so on...

4/- First give us reliable systems ( and I mean systems that are not designed for failures - as you all claim they are - because someone like me is going to be there to mind the store when the chips are down...), then give us pilot-friendly interfaces that have eliminated all erroneous interpretations and I, for one, would consider that, maybe, we could go a bit further.
That bit further is certainly not single plot operation. because of the psychological problems arising from being locked alone for hours in a cramped flight deck...ask the psychologists first !

But as Zappbrannigan says, it could happen in a hundred years, I can't care less as I'll probably be "treading the untrespassed sanctity of space by then  

Quoting Rheinwaldner (Reply 213): Quoting Tdscanuck (Reply 207): It's actually relatively trivial...it's just a SIMO controller that takes the desired yaw rate as input and drives the rudder, thrust, spoilers, nose wheel steering, and brakes as outputs. This type of control problem is extremely well studied and not all that difficult...there are much more challenging issues facing a pilotless airliner. How about the optimum configuration? Could be all gears in, but if the danger is extreme break away to one side (e.g. because the front wheel is stuck at an 45° steering angle) an idea could be to extend only the main gear on the other side.

If that's actually the best solution, why wouldn't you just program that in? You've got (multiple) nose gear angle sensors. The issue is the unknown failures...known failures with known solutions aren't really that tricky.

Quoting Rheinwaldner (Reply 213): Quoting Tdscanuck (Reply 207): For the Turkish and LX cases, I would have to assume that any pilotless airliner would include suitable landing areas, not just active airports, in the nav database. For LX: The airport was not just inoperative in was actually broken. A wall of mud crossed the runway. It was enough suitable for the passengers to survive. It would not have been in any database. For TK: Problem was loss of any navigation system. What use has any database in this case?

"Suitable landing areas" should be anything that can take a landing. Runways, dry lake beds, etc. I'd actually be more surprised if they kept the database up to date enough to know about the wall of mud than if they didn't have it in the database.

For the TK case, it has to be a baseline assumption that the navigation system (position, not autoflight) has at least the same level of reliability as the flight controls, since it goes from being a "very nice to have" to an "absolutely essential" for safe flight.

Quoting Rheinwaldner (Reply 213): I come to the conclusion that most of today emergency landings require the presence of pilots onboard. In most cases the reason for emergency landings are failed systems. For fully automated planes therefore a paramount improvement in system MTBF must be achieved. Otherwise nearly any of todays emergency landings would be crashes.

Most of today's emergency landings are caused by equipment failures, but most (the *vast* majority) are non-safety-of-flight failures, so why would they end in crashes? An engine failure often results in an emergency landing, but an engine failure is only considered an economic (not safety) event.

Quoting Pihero (Reply 218): 1/- That event occurred sixty five years after very little experienced youngsters managed to bring back home a few "un-flyable" bombers which have suffered horrific damage.

True. However, given that those were totally manually controlled aircraft designed for combat, I'm pretty certain they had a great deal more static stability than current aircraft (both because of design conservatism and to sustain battle damage).

Quoting Pihero (Reply 218): 2/- An F-18, like all modern *combat* aircraft are built both for manoeuvrability and for damage tolerance, to the point that it is quite difficult to know where the roll / yaw /pitch controls start and where they end. On any airliner you can think of, you'd have dedicated controls... and unless you design future generations of air transport airplanes with that goal, (and why not, apart from the weight and cost considerations, which are at least very important ?) the demonstration is not valid.

I'm not sure why it's invalid...it was done on an aircraft that's *far* more difficult to control than an airliner, so the problem should be less of an issue on airliners than fighters. On current design airliners there's already considerable control cross-coupling (notably aileron-rudder and aileron-spoiler) and the A350/787 are taking this considerably farther.

Quoting Pihero (Reply 218): Now, for the proponents of "all-automation-forward", this little link about the X-43 which is as far as I understand sponsored by both NASA and DARPA : Cause of X 43 loss, from which I extract these gems : "...the failure of the NASA X-43A hypersonic aircraft in June 2001 was the result of inaccuracies in computer and wind-tunnel tests that were based on insufficient design information about the vehicle itself... According to the MIB documents, the X-43A Hyper X launch vehicle "failed because the vehicle control system design was deficient for the trajectory flown due to inaccurate analytical models (Pegasus heritage and Hyper X launch vehicle specific), which overestimated the system margins... The X-43A mishap board points to a series of "modeling inaccuracies" used in ground tests, as well as "misinterpretationn" of wind tunnel data due to insufficient data. ... The board found that the flight mishap could only be reproduced when all of the modeling inaccuracies are taken into account in post-failure work dedicated to unraveling the cause of the X-43A mission loss....

Quoting Pihero (Reply 222): 2/- That example showed -and that was the point- that the design was faulty not to a lack of data but to a lack of having considered all the factors, and (quote)an overestimation of the margins(unquote). It means that humans, be they pilots and engineers are susceptible to errors (in this case by omission).

I'm not clear on how this was an automation failure...if you screw up the control modelling for any control system (that isn't adaptive) both the computer and the human pilot will crash.

Quoting Pihero (Reply 218): I think that you underestimate the problem : 1/-Nose wheel steering is lost 2/-Rudder could be effective but will lose that effectiveness under some 60 knots. 3/- If you use the brake and the spoilers ( btw these are still symmetrical as ground spoilers) for directional control, you lose them effectively as deceleration devices...and as reverse thrust won't be that efficient, would you consider using forward thrust to counter the yawing moment of the nosewheel ?...... So let's see : Loss of half the braking power + one engine in high forward thrust + no spoilers =....runway overrun anyone ?

The assumption for a fully integrated control system like that is that you have independant control (the "MO" of "SIMO") of the actuators. So have differential braking (which is available today), differential thrust (also available today, in both forward and reverser), differential spoiler (available today), etc.

Quoting Pihero (Reply 218): What I also see is that WE COULD design a fully automated airliner... but "could" is not "can safely" and so far to all the arguments Max, Rheinwaldner and others have said, the answer has been mainly " more sensors, more info" which , in our to-days modern technology leads to a massive complexity of the system,a lot of weight increase

I agree there's complexity and cost barriers, and that's it's technically possible. It's not clear that it would be less safe, nor it is it clear that it would be safer (mostly because we don't have enough data on the current state and we don't know yet know what it would be like if we designed for pilotless from the start). However, I'm not sure the weight increase is actually true. You'd certainly gain weight in sensors and wiring...probably not in computing since computer power isn't well tied to size.

However, in return for that you'd have huge structural weight savings by removing the windows and associated structure, flight deck, flight crew support systems, and the flight crew themselves. You'd easily have a budget of a couple of tons to work with before weight started becoming a problem.

Quoting Pihero (Reply 218): for instance, I'd really like to see hundreds of aircraft sharing the same airspace, each one radiating radar pulses, LIDAR waves plus two dimensional cameras looking around all sectors around the aircraft....let's be serious, our finance managers would have a fit at the cost, plus the one of hiring countless of new engineers and technicians ( which is still in my mind the whole subject of this thread)...

There's certainly a capital expenditure hike, but one aircraft requires multiple flight crews...how much does an airline spend on all the flight crews over the ~20 year life of the aircraft? I bet that's a pretty hefty number too.

Quoting Pihero (Reply 222): I wanted to show that what was achieved by Rockwell-Collins required a totally different flight control architecture than is usal to all airliners I know.

But if we do a pilotless airliner, I think it's a safe assumption that it's going to have a totally different flight control architecture anyway, so I'm not sure why this is an issue.

Quoting Pihero (Reply 222): 1/- Automation so far is only good for procedural problems and certainly not able to cope with what Tom and Rheinwaldner call "soft decisions", which are the major part of flying an airplane.

It's definitely bad at *some* "soft decisions", but not all. That's one of Bond's main points, as I see it...automation can do a lot more than "if-then."

Quoting Pihero (Reply 222): In the example of the nose wheel stuck at a 45° angle from straight is also telling : The safest solution is a belly landing, as all others are linked to a degradation of the airplane performance -and evacuation-. Problem is : that solution requires a certain amount of lateral thinking and considerations outside the technical side, which engineers don't do generally in this kind of situation.

As with the top of this post, if we know what the safest solution is, that seems like a trivial issue. Anything that easily and reliably goes in the QRH is easily automated.

Quoting Pihero (Reply 222): MELed item : APU inop...aircraft dispatched to a place in North Africa where the temperature is 40° C +, prevision of 185 passengers on the return leg. I called both dispatch and engineering : I'm not taking that airplane there as there won't be any ground conditioning cart available and that temp is going to cause havoc on my passengers, my crew, my electronics, my performances...They came quickly to my conclusions and the flight was delayed for two hours for an airplane swap.

I don't think anybody is proposing humanless dispatch yes.

Quoting Pihero (Reply 222): 2/- Replacing pilot mistakes by engineering design errors or oversights ( the result is the same : a faulty system) is not exactly what I would call progress.

It may not be progress on safety at all. However, even if it held even, there are advantages (and disadvantages) on other fronts. The balance is the key, which I think you and I agree on.

Quoting Pihero (Reply 222): 3/-Contrarily to what some people are saying, we have reached a plateau on flight deck automation and I can't really think of any change since the 744 which flew nearly thirty years ago...Better communications, perhaps, better flight controls ( à la Airbus A380) and what else ?

There's a huge reduction in flight deck controls from the 747-400 to the 777 to the 787...the 787 is doing a lot more automatically than the 747-400 did. I'm not sure if that's a plateau, but it's certainly not "no change."

Quoting Pihero (Reply 222): 4/- First give us reliable systems ( and I mean systems that are not designed for failures - as you all claim they are - because someone like me is going to be there to mind the store when the chips are down...)

All systems are designed for failures...that's the entire purpose of FMEA which is applied, in some form or another, to pretty much everything on the aircraft. I think you means "designed *to fail*", which isn't actually what I said.

No system is designed to fail (fail = perform other than as intended). What I appear to be having a great deal of time cogently explaining is that some systems allow certain failure modes *because* there is a suitable backup present. That may be another system, a human, etc. When you get into failure mitigation, you always need to look at the importance of the system, the causes and effects of the failure, and the cost of failure mitigation. For example, autoflight systems aren't all that reliable. It's possible to design considerably more reliable autopilots, at significant cost. But there's no point...if the autopilot dies, there's a perfectly competent flight crew right there who can do everything the autopilot can do. So there's no reason to spend the money to crank up the autopilot reliability beyond that required for economic operation. In contrast, the engines (as a system) have to have *much* higher reliability, and the money is spent to achieve that, because there's no backup.

Tom.

Quoting Tdscanuck (Reply 223): If that's actually the best solution, why wouldn't you just program that in?

Because there are so many failures that haven't been accounted for. Just take any FCOM and see - on this subject - the very small number of forecasted landing gear / tyre failures.
I do not think that you can guarantee that you've accounted for every possible failure on an aircraft.

Quoting Tdscanuck (Reply 223): I'm not clear on how this was an automation failure...if you screw up the control modelling for any control system (that isn't adaptive) both the computer and the human pilot will crash.

My point is that designers can screw up, too. The question is :"Are their mistakes more acceptable than a pilot's to the point that they can design the pilot's replacement ?"

Quoting Tdscanuck (Reply 223): The assumption for a fully integrated control system like that is that you have independent control (the "MO" of "SIMO") of the actuators. So have differential braking (which is available today), differential thrust (also available today, in both forward and reverser), differential spoiler (available today), etc.

Granted, but all lead, as I said to degraded modes of operations of each one of these systems, to the point that in all likelihood, the result is a crash at the runway end...if they ever managed to keep the airplane on the centerline, which I doubt very much.

Quoting Pihero (Reply 222): As with the top of this post, if we know what the safest solution is, that seems like a trivial issue. Anything that easily and reliably goes in the QRH is easily automated.

But the first condition of that success is to have thought of the failure...which we have agreed is not the case, generally as problems have a nasty way of revealing themselves when no-one has foreseen them.

Quoting Tdscanuck (Reply 223): I don't think anybody is proposing humanless dispatch yes.

And why not ? After all, they deal with an array of known facts, known alternative solutions to a finite set of *classic* events ( loss of pressurisation, terrain overflight, weather at alternates) to an uneventful flight... The unknown is left to the captain to be concerned with.

Quoting Tdscanuck (Reply 223): There's a huge reduction in flight deck controls from the 747-400 to the 777 to the 787...the 787 is doing a lot more automatically than the 747-400 did. I'm not sure if that's a plateau, but it's certainly not "no change."

Qualitatively,I agree with you : in-flight electronic libraries / data links / ACARS... but they haven't changed my job, have they ? The info they give is faster, may be mor accurate, but I still manage the flight the way I used to in the eighties on the 1011, with exactly the same navigational monitoring, the same inputs to the NAV systems...etc...

Quoting Tdscanuck (Reply 223): I think you means "designed *to fail*", which isn't actually what I said.

I beg your pardon, that's what I meant...English is not my native language...

Quoting Tdscanuck (Reply 223): What I appear to be having a great deal of time cogently explaining is that some systems allow certain failure modes *because* there is a suitable backup present.

Tom, this is one of the very points on which we disagree : the two failures that have been discussed - the Malaysian T7 and the Spanish 320 - looked at from that design point are akin to criminal intents, when they are only design oversights in my opinion.

There is a NASA document that projects the cockpit environment into the planned NGATS - New Generation Air Traffic System, for 2025 - called "The Intelligent Integrated Flight Deck[" is the paper # 4 of the readings on CAFE, the link of which is at the bottom of this post. It is worth reading as it is at the heart of our discussion. For the automation proponents, just replace *human* by *computer* and we'd see the beginning of the problems involved.
The introduction tells the scope of the project :
"Nasa, NTSB...CAST...AOPA...have recognized that pilots, automation, and the external environment continue to be causal factors in a large percentage of accidents/incidents. Transitioning toward the NGATS must not only address these findings, but also the implications on flight deck system safety when considering new operational capabilities. Specifically, the envisioned environment includes more automated functions, increased expectations of the flight deck crew, less time to "see-and-avoid", and increased system complexity. Each of these implies a need for flight deck system safety research."...

On page 6 there is an already very interesting diagram on the scope of the project.
Hope we can discuss it.

Quoting WPIAeroGuy (Reply 221): there was an article I pointed out earlier in which the eCFI as it was called noticed the aircraft was drifting off course. It aurally alerted the pilot then activated the stick shaker, but there were no control forces on the stick so it assumed that the pilot was unconcious. It then selected a suitable alternte, contacted the tower, and made an emergency landing to awaiting rescue services. And, this entire system was created by a retired electrical engineer.

Since you mentioned it, I couldn't help feeling that the story was too pretty to be true and I had a hard look at the net...And then, I found this article from which this is an except :
"7.5. Sample Human Non/Rare-Normal Scenarios
7.5.1. Scenario Title: Pilot incapacitation
The Actual system displays guidance to the pilot and signals the pilot when to perform the maneuvers. If
the pilot fails to perform, the Actual system will make concerted efforts to engage the pilot's attention. If
the pilot is still not responding, the Actual system will announce a potentially incapacitated pilot and fly
to the nearest airport (note that this is not necessarily the alternate that pilot may have selected). It will
not take facility or transportation considerations into account. This would simply be the fastest way
down. The H will perform the landing on its own, unless the pilot tries to intervene. If the pilot
intervenes, the H will return to its normal operations."
Doesn't that sound like your mentioned article ?

The whole paper from NASA, "The naturalistic Flight Deck System"

is # 7of the proposed readings on the main project :

The eCFI Electronic Certificated Flight Instructor

Please note that all these papers deal with flight deck automation improvements and all the diagrams, studies show that - at least until the year 2025 horizon - there is still no scope for a successful autonomous aircraft.
May I rest may case for the moment, your Honor ?

Quoting Pihero (Reply 224): Quoting Tdscanuck (Reply 223): If that's actually the best solution, why wouldn't you just program that in? Because there are so many failures that haven't been accounted for. Just take any FCOM and see - on this subject - the very small number of forecasted landing gear / tyre failures. I do not think that you can guarantee that you've accounted for every possible failure on an aircraft.

Honestly, until you can grasp the concept that computer systems do not need to be 'programmed' for every single specific combination of conditions and failures that can ever occur, but can still provide the correct responses, it really is difficult to have an intelligent discussion on the subject.

Jimbo

Quoting Pihero (Reply 224): Quoting Tdscanuck (Reply 223): If that's actually the best solution, why wouldn't you just program that in? Because there are so many failures that haven't been accounted for. Just take any FCOM and see - on this subject - the very small number of forecasted landing gear / tyre failures. I do not think that you can guarantee that you've accounted for every possible failure on an aircraft.

I think we can guarantee that we will *not* account for every possible failure on an aircraft. I was just pointing out that, for any failure we can foresee, we can design a procedure to deal with it because that's what we do today.

It's only failures that weren't anticipated that will potentially cause meaningfully divergent responses between a human and a computer.

Quoting Pihero (Reply 224): Quoting Tdscanuck (Reply 223): I'm not clear on how this was an automation failure...if you screw up the control modelling for any control system (that isn't adaptive) both the computer and the human pilot will crash. My point is that designers can screw up, too. The question is :"Are their mistakes more acceptable than a pilot's to the point that they can design the pilot's replacement ?"

I don't think they're more of less acceptable...if we ever did attempt a pilotless airliner, we'd have to be considerably more dillegent about the design and testing for exactly this reason.

Quoting Pihero (Reply 224): Quoting Tdscanuck (Reply 223): The assumption for a fully integrated control system like that is that you have independent control (the "MO" of "SIMO") of the actuators. So have differential braking (which is available today), differential thrust (also available today, in both forward and reverser), differential spoiler (available today), etc. Granted, but all lead, as I said to degraded modes of operations of each one of these systems, to the point that in all likelihood, the result is a crash at the runway end...if they ever managed to keep the airplane on the centerline, which I doubt very much.

But if you've degraded the operation of all the systems to the point that the airplane can't stay on centerline, it doesn't make a difference if you've got a human or a computer...physics are against you at that point. If the aircraft is theoretically controllable, a properly designed control system can do a better job than a pilot for aircraft control simply because it can react much faster and command a much wider range of simultaneous control inputs. This is very well demonstrated on unstable fighters like the F117 or F22.

Quoting Pihero (Reply 224): Quoting Tdscanuck (Reply 223): I don't think anybody is proposing humanless dispatch yes. And why not ? After all, they deal with an array of known facts, known alternative solutions to a finite set of *classic* events ( loss of pressurisation, terrain overflight, weather at alternates) to an uneventful flight... The unknown is left to the captain to be concerned with.

I would argue that the biggest reason is that the array of known facts is far larger for the dispatcher than the pilot...there's certertainly one, probably two or three, orders of magnitude difference.

Quoting Pihero (Reply 224): Quoting Tdscanuck (Reply 223): What I appear to be having a great deal of time cogently explaining is that some systems allow certain failure modes *because* there is a suitable backup present. Tom, this is one of the very points on which we disagree : the two failures that have been discussed - the Malaysian T7 and the Spanish 320 - looked at from that design point are akin to criminal intents, when they are only design oversights in my opinion.

Actually, I think we agree on this. FBW is one of those systems where most failure modes should *not* be allowed because there is no credible backup (especially on widebodies). Systems where a failure mode may be acceptable are things like autopilot or FMC where failure does not jeopardize safe operation of the aircraft because the flight crew is capable of executing those same functions.

Tom.

Quoting Tdscanuck (Reply 226): I would argue that the biggest reason is that the array of known facts is far larger for the dispatcher than the pilot...there's certainly one, probably two or three, orders of magnitude difference.

Having worked both as a line pilot and as such, as a member of an OCC/Dispatch nerve center, I can guarantee that your statement is false...up to the information service that can be provided to an aircrew in flight : medical emergency for instance : nearest place for treatment taking the urgency into account...airspace closure : new strategies for a re-routing...etc...(we do not use the US method of semi-remote control...the captain retains ultimate decision and authority).
Furthermore, that information is already in many cases automatically provided to the aircrews.
Order of magnitude ? Where ? and on that very subject of *knowing facts*, isn't that exactly what the computers are good at ?

Quoting Bond007 (Reply 225): Honestly, until you can grasp the concept that computer systems do not need to be 'programmed' for every single specific combination of conditions and failures that can ever occur, but can still provide the correct responses, it really is difficult to have an intelligent discussion on the subject.

Yes, for as long as you only concentrate on technical proceduring (may I use that word ?)...
You can have adaptive flight controls, better sensors..You still have not demonstrated that outside the technical set-ups you can cope with every event that happens on an aircraft.
See all the scenarios we pilots have proposed and to which we do not have, from you, the beginning of a solution.
What's the use of superfast computers when they can't deal - by inherent design impossibilities, so far - with events outside the "piloting" task of flying the airplane ?
Read that NASA paper on the intelligent system and see if really their engineers have come up with an automated solution (the answer is no, they can't...yet, and their solution is to provide a far better human environment...and that for the year 2025, which tells a lot about their encountered problems and their optimism for finding a quick solution to an all automated airliner.)
So, Mr Bond, the pilot part of you can't still solve my little riddle ?

And now, to show that I'm quite open, this paragraph from Tom :

Quoting Tdscanuck (Reply 226): I think we can guarantee that we will *not* account for every possible failure on an aircraft. I was just pointing out that, for any failure we can foresee, we can design a procedure to deal with it because that's what we do today. It's only failures that weren't anticipated that will potentially cause meaningfully divergent responses between a human and a computer.

and to which I heartily agree. The question is - and remains - which one is going to be consistently better at decision making outside the procedural field ? Of course the loser will be the other one's slave.

Quoting Pihero (Reply 224): I do not think that you can guarantee that you've accounted for every possible failure on an aircraft.

It is a requirement to demonstrate that you have accounted for every foreseeable failure.

Quoting Pihero (Reply 224): Are their mistakes more acceptable than a pilot's to the point that they can design the pilot's replacement ?

Yes. They have five to eight years to exhaust all alternatives and tangents, have decades of knowledge available for consultation, have extensive simulation capability to validate choices.

This is why aircraft systems have progressively taken over pilot responsibilities and will continue to do so.

Removing the pilot from the cockpit could be done today - from a purely technological point of view. The limit is public acceptance and government regulation. Evolutionary development will, in time, make it a very plausible possibility even with these limitations.

Quoting Bond007 (Reply 219): That's exactly how you described the 'human' scenario earlier ... there is a smell in the the galley, and the F/A made a decision on whether it was burnt coffee, or burnt wiring insulation, and forwarded that information to the cockpit ... whether there is a human or a computer behind the cockpit door is totally irrelevant.

You really promote the "divert immediately" button for the cabin crew? How can F/A's find out whether a certain alternate looks not so promising because the weather gets worse or traffic congestion at one alternate would make another more preferable? How? How do they know about snowstorms, runway restrictions, closing airports etc.?

Quoting Tdscanuck (Reply 223): If that's actually the best solution, why wouldn't you just program that in?

Good lock with the use-cases of that program! And good luck with the acceptance testing of that feature! You have thought about testing of system features? If it is a feature it MUST be tested! How would a test series look like? Ten landing attempts each with a front gear that is arrested at different angles by 10 degree steps?

Quoting Tdscanuck (Reply 223): Quoting Rheinwaldner (Reply 213): I come to the conclusion that most of today emergency landings require the presence of pilots onboard. In most cases the reason for emergency landings are failed systems. For fully automated planes therefore a paramount improvement in system MTBF must be achieved. Otherwise nearly any of todays emergency landings would be crashes. Most of today's emergency landings are caused by equipment failures, but most (the *vast* majority) are non-safety-of-flight failures, so why would they end in crashes? An engine failure often results in an emergency landing, but an engine failure is only considered an economic (not safety) event.

Ok, you are right, the majority could be handled by an automated system. But the rest in absolute numbers surely still exceeds the number of crashes by pilot error.

Quoting Pihero (Reply 224): Quoting Tdscanuck (Reply 223): I don't think anybody is proposing humanless dispatch yes. And why not ? After all, they deal with an array of known facts, known alternative solutions to a finite set of *classic* events ( loss of pressurisation, terrain overflight, weather at alternates) to an uneventful flight... The unknown is left to the captain to be concerned with.

  
About automation candidates: IMO dispatching is a far better one than pilotless planes.

Quoting Bond007 (Reply 225): Honestly, until you can grasp the concept that computer systems do not need to be 'programmed' for every single specific combination of conditions and failures that can ever occur, but can still provide the correct responses, it really is difficult to have an intelligent discussion on the subject.

If you promote computers in AI operation mode to control aircrafts you took the cake! This is bloodcurdling! Mankind struggles to make software do what is expected, but you want to implement that kind of control for aircrafts? You would earn multitudes of erroneous reactions = crashes. An automated aircraft system needs an extreme testing effort. How do you want to test unforeseen responses? It does not work, never. I was never so confident using that word.

Quoting Tdscanuck (Reply 226): It's only failures that weren't anticipated that will potentially cause meaningfully divergent responses between a human and a computer.

How could that gap be ever closed? It would require to anticipate every condition, also those that have never occurred before.

Quoting Tdscanuck (Reply 226): I don't think they're more of less acceptable...if we ever did attempt a pilotless airliner, we'd have to be considerably more dillegent about the design and testing for exactly this reason.

I want to stress that point, it is of course correct. IMO full scale testing of automated airliners would result in three to five hull losses per aircraft type. Otherwise you won't get that aircraft tested. In software development there are terms like crash-test, smoke-test, stress-test. To test the demanded scope of functionality you have to go beyond it.

I have a firm believe in technology, automation and computer stuff. You should see how my house is automated (by self written software). Yet I am convinced that replacing pilots is such a bad automation candidate.
Just to show my sympathy for extravagant ideas here the link to my aerial refueling thread:
https://www.airliners.net/aviation-fo...ums/tech_ops/read.main/226906/1/#1

In this thread in post 11 I mentioned the possibility of automated refueling. This would overcome a lot of the concerns that were risen (crew training, make aerial refueling easy and safe). And such an automation would be much easier than banning pilots from a plane. Especially if auto-flight functionality is so highly developed as revealed in this thread.
I am still convinced that the aerial refueling concept would save enormous fuel. That is what I would call a promising automation candidate. Both in return as in feasibility.

I have another two aviation ideas which I haven't expressed so far in publicity. They are so "unrealistic" that I hardly dare to mention them. But with a fraction of automation and accurate flight path control than what would be required by a pilotless plane these ideas suddenly become feasible. In other words: with the level of technology required by pilotless airplanes much bigger revolutions could happen.

Quoting Rheinwaldner (Reply 229): You really promote the "divert immediately" button for the cabin crew? How can F/A's find out whether a certain alternate looks not so promising because the weather gets worse or traffic congestion at one alternate would make another more preferable? How? How do they know about snowstorms, runway restrictions, closing airports etc.?

No, that's not what I said. I was explaining how the initial response to this scenario, is totally independent of what/who is behind the cockpit door. The scenario, as described, was that there was a smell of smoke in the galley, and the F/A was the one that made a decision on whether it was burnt wiring insulation, or burnt coffee. This is simply a 'problem' or 'no problem' input to the cockpit .... and the human or computer system will make a decision based upon that input. Snowstorms, traffic ..... this is no different than any other automated scenario where there is a snowstorm or traffic! No less data is available just because a F/A signalled that there is a problem.

Quoting Pihero (Reply 227): You still have not demonstrated that outside the technical set-ups you can cope with every event that happens on an aircraft

LOL ...I didn't realize we were here to demonstrate that a system can cope with every single event that happens on an aircraft. We've explained why we don't need to ... and why a system doesn't have to be programmed with a list of all possible scenarios that can ever happen.

The major misunderstanding here, is the thinking that without a list of all the millions of possible scenarios somehow programmed into the computer, it cannot cope with unexpected situations. This is just as illogical as saying that unless a human has been taught how to deal with every possible scenario, he/she will have no clue how to control the aircraft when something unexpected happens.

Quoting Pihero (Reply 227): So, Mr Bond, the pilot part of you can't still solve my little riddle ?

  

Quoting Pihero (Reply 227): See all the scenarios we pilots have proposed and to which we do not have, from you, the beginning of a solution.

Actually, you've had answers to every single question you've asked ... you just either do not accept the answers, or in most cases, do not understand (or wish to understand), how automated computers systems can and could work.

Quoting Rheinwaldner (Reply 229): How do you want to test unforeseen responses? It does not work, never.

Well, the question is valid today ... nothing is different. Aircraft engineers have the same challenges with every aircraft they design.



Jimbo

Quoting Bond007 (Reply 230): The major misunderstanding here, is the thinking that without a list of all the millions of possible scenarios somehow programmed into the computer, it cannot cope with unexpected situations.

The whole day at my workplace I struggle with computers that do one thing: cope with unexpected situations, just ... they fail miserably. Computers always stumble over unexpected situations. Computers never execute one command that has not been programmed. Please dismiss the idea that computers could ever work in artifical-intelligence mode in an airplane, it hurts so much. Why:

Aircraft automation must make use of software that is written after "DO-178B": This standard requires amongst many other things:
"One of the key concepts behind DO-178B is that all the requirements are covered by the code and that all the code that exists is directly due to a requirement."
This implicates that any line of code is coupled to a requirement. It categorically excludes anything like not delibertaly programmed responses.

Read here:
http://www.telelogic.com/Standards/do-178b.cfm
http://www.lynuxworks.com/solutions/milaero/do-178b.php3
http://www.esterel-technologies.com/...les/AeronauticsHandBook-DO178B.pdf Chapter 2.4

For airplane software that has catastrophic consequences in case of failure the burdens are so high that you don't want to put in too much functionality at all. About the Airbus flight control system I heard around 1999 that the initial certified code from the eighties was still unchanged. The effort to get that piece of software done was tremendous. In the software industry that piece of code is considered as one of the most high-tech things.

Quoting Bond007 (Reply 230): This is just as illogical as saying that unless a human has been taught how to deal with every possible scenario, he/she will have no clue how to control the aircraft when something unexpected happens.

That is the difference between man and computer. In having a clue man supersedes any computer. The clue the computer has never goes beyond the clue of the software engineering team.

Quoting Pihero (Reply 224): Doesn't that sound like your mentioned article ?

That's not the article. The article I read was either in Flight Training Magazine, or Sport Pilot magazine. When I go home this weekend I'll get the magazine. And it was definely called the e-CFI.

Quoting Rheinwaldner (Reply 229): Good lock with the use-cases of that program! And good luck with the acceptance testing of that feature! You have thought about testing of system features? If it is a feature it MUST be tested! How would a test series look like? Ten landing attempts each with a front gear that is arrested at different angles by 10 degree steps?

If a piloted airliner doesn't have to be tested for it, why would a pilotless one? Simulators can show that the software executes correctly, and as long as the airframe performs during flight tests, I see no reason why you'd have to demonstrate every failure in real life. I am NOT saying that certification wouldn't be different, it would be far more stringent, orders of magnitude more.
Pilots practice in simulators, why can't an autonomous pilot go in a simulator?

Think about it this way...all of the examples of why autonomous airliners couldn't work people have brought up. There's like what, 5 or 6 of us on a message board? Imagine a team of 100 engineers working 50 hour weeks, there sole job to come up with every possible scenario. The issue with the blocked runway, it wouldn't be in a database, but then how would a pilot deal with it? They'd talk to the tower. The tower can easily communicate to an airplane that runway X is unavailable, whether its autonomous or not.

Another idea we've skipped over is the networking of these aircraft. Each one could learn, in real time, from every other airplane around the globe. For instance, based on the weather radar, airplane X decides to take a certain route. Accelerometers measure turbulence, and determine the ride is too rough for passengers, and a new route is determined. This decision is then broadcast to all other autonomous aircraft, so when faced with similiar situations they're decision making is slightly altered.

And around we go again.
Perhaps the message will be better understood if it came from THE specialists in this area...NASA !
This is just a piece of the paper I gave the link to ( #7) on a previous post.
See that what it says sums up just about everything we have said and not the mantra "if you could do it, a computer will do it better" :
"...We live in a world where errors are made by humans and errors are corrected by humans. Humans make errors in design, manufacture, direction, operation, and maintenance. At every stage of implementation and execution, humans can introduce errors and mitigate them.
Conventional thinking claims that removing the human interaction and replacing it with machines will remove human error. However, this thinking overlooks two important considerations. The first is that humans are involved at some point in virtually every human endeavor, so it is nearly impossible to completely remove the human (e.g., removal of the human from operations via automation will simply move human errors to the design phase). The second consideration is perhaps more meaningful—the human’s ability to prevent and mitigate unforeseen contingencies or situations.
Recent history is replete with examples of design, manufacturing, and maintenance errors causing severe failures (e.g., electrical power grids, airline and airport dispatch operations, and autonomous vehicles) that could have been prevented had there been the opportunity for some form of human intervention. In most of these cases the designers, manufacturers and maintenance personnel “failed to anticipate…” or “did not consider…” some combination of factors or occurrences.
This situation is not surprising given that in complex systems there are essentially an infinite number of potentially significant factors and interactions. If a human had had some level of access to the system during operation, he might have performed a simple action, such as isolating a section of the power grid or correcting for an erroneous piece of data, that could have prevented or mitigated the failure. While supporting such access is often rejected based on statistics implicating the pilot as the top causal factor in current accidents, it is extremely important to recognize that no comparable data is recorded on successful human interventions (indeed, we are not sure how such data could be collected), that is, the number of failures, errors, and ultimately accidents that are avoided or remedied by human actions. Without data on beneficial interventions, using accident statistics to advance the case that human pilots “usually” make things worse and thus their system access should be minimized or eliminated is misleading and would be like saying that because engine failures lead to crashes, engines cause aircraft to crash and should be eliminated. Based on such flawed logic, traditional automation philosophy tends to throw the baby (human prevention and mitigation abilities) out with the bath water (human error). Complemation (Schutte, 1999) or complementary automation offers a better alternative. Complemation does not remove the human from the operational aspects of the mission; rather it takes advantage of the human’s unique abilities. Complemation deals with errors that the human pilot could potentially introduce into the system by reducing the proneness of the system to human error, identifying human errors and making them apparent to the pilot, and providing prophylactic measures to insure that human errors do not propagate throughout the system causing a failure. Thus, the machine and the human work together in symbiosis. The human pilot provides common sense knowledge, general intelligence, and creative thinking while the machine provides swift and precise control, extreme vigilance, resistance to fatigue, and encyclopedic memory. The human pilot has many tasks but perhaps the most important one is to compensate for limitations of other parts of the system or put more simply, to handle the unexpected...."

What else have we said ?

BTW, the rest of the article, and in a matter of fact the whole bunch of texts in the CAFE site are worth a good look at.
Maybe some will understand what a pilot's job entails a bit more accurately and some so-called pilots will have a good look at the way they perform their job and maybe choose another line of trade. They don't belong in a flight deck. I certainly do not want them in mine.
Finally, there is a thread in thechlog of a prune net that deals with the human/black boxes relationship. Might be entertainig too.

[Edited 2008-10-23 16:16:45]

Quoting WPIAeroGuy (Reply 232): That's not the article. The article I read was either in Flight Training Magazine, or Sport Pilot magazine. When I go home this weekend I'll get the magazine. And it was definely called the e-CFI.

The article I quoted is part of a series of 7 papers under the title of "The eCFI", of which the excerpt is a NASA study dated December 2007.
I honestly can't believe that a project on which a crowd of specialists are still trying to define the scope have been beaten on the exact scenario they envisioned by someone working from the corner of his kitchen table, realised it and then had it flown by someone who generously had a heart attack in order to be saved by the machine.
I may be innocent but not that gullible.

Reinwaldner,

You continue to bewilder them with facts, your well thought out, logical argument with the benefit of extensive experience in computers makes no difference to this small group of people obsessed with this concept no matter how little it makes sense.


Mr Bond seems to have some experience with computers and extensive experience of nothing else as is apparent from his refusal to detail any aspect of his 'thousands of hours of flight time'


He cannot answer Pihero's questions because he is unable, his credibility is non existent.



So we have a small group of people who, exclusively have no Pilot qualifications or experience
in any real life use of their beloved computers in the cockpit or making the thousands of decisions covering myriad scenarios 24 / 7 / 365 days a year proposing all of these can be addressed autonomously.

When ridiculous ideas like letting Flight Attendants make decisions on whether or not to divert in life threatening situations are actually put forward as serious solutions most of us realise how much this group has totally lost touch with reality.

Pihero: Yes, human errors occur in the design phase as well. But whereas design glitches can be tested, corrected and overridden before the technology goes into service, a pilot's error is permanent and irrecoverable.

'A Pilots error is permanent and irrecoverable'

Interesting statement, if any of us make a mistake in the cockpit when we do not catch it ourselves we have another (human)
independent monitor to correct us and or suggest a better solution or course of action.

This is the cornerstone of Crew Resource Management and is practiced on a daily basis.

In fact the direct opposite of 'permanent and irrecoverable' !

Pihero,

I agree with everything that article said. None of us have ever said lets remove all humans, we're just removing the pilot from the airplane. The problem with the anti-autonomy arguement you are all makingis that it assumes a 1-1 transition of pilot for computer. It won't be like that. Flight attendants will need additional training to become a better interface for what happens in the cabin and what the flight computer decides to do.

Look at the fundemental idea here: move people, via an airplane, from one place to another without direct operation of the airplane by a human. By changing the entire infrastructure of commerical aviation it is most certainly possible. Everything now is designed for humans to operate. Thats why there are lights and painted lines on an airport. It would much easier to use GPS and perhaps and embedded EM path for computer navigation, because currently computers can do that better than humans.

Quoting Max Q (Reply 237): This is the cornerstone of Crew Resource Management and is practiced on a daily basis.

Your right, and 99.99% of the time it works perfectly. There are times, and crashes have shown that it doesn't always. This goes back over a hundred posts to when I said that computers don't make stupid mistakes, they don't have brain farts. Like Tds has said numerous times; like Pihero's article says, no one knows who saves who from more mistakes, computer or human. That's not a case for or against automation, its simply a statistic that would have to be more seriously studied should an autonomous airliner see the light of day.

Quoting Max Q (Reply 235): When ridiculous ideas like letting Flight Attendants make decisions on whether or not to divert in life threatening situations are actually put forward as serious solutions most of us realise how much this group has totally lost touch with reality.

Brilliant argument. Think outside the box a little. Its an infrastructure change, not a robot who sits in the left seat. You can't think past the instrument panel, and are so consumed by your own convictions that you are unable to grasp any other facts that people are pointing out.

Rheinwalder has made some very good arguments, and I'd say he is the only legitimate debater against automatino. However, just because he is a programmer who think its can't be done doesn't mean it can't. Frankly, none of us are experts at autonmous aircraft, and we all know just enough information to cause a debate that most likely will never end. I think everyone needs to keep that in mind.

Oh I found the article. It was in Kitplanes which is why I couldn't find it. The article can be found here for subscribers:
http://www.kitplanes.com/issues/24_6/builder_spotlight/7886-1.phtml

Quoting Pihero (Reply 227): Quoting Tdscanuck (Reply 226): I would argue that the biggest reason is that the array of known facts is far larger for the dispatcher than the pilot...there's certainly one, probably two or three, orders of magnitude difference. Having worked both as a line pilot and as such, as a member of an OCC/Dispatch nerve center, I can guarantee that your statement is false...up to the information service that can be provided to an aircrew in flight : medical emergency for instance : nearest place for treatment taking the urgency into account...airspace closure : new strategies for a re-routing...etc...(we do not use the US method of semi-remote control...the captain retains ultimate decision and authority).

Hmm...maybe what I'm thinking of as dispatcher and what you're thinking of are differet things. Or I may have misworded what I wrote..."knowable facts" would have been more accurate than "known facts."

Quoting Pihero (Reply 227): See all the scenarios we pilots have proposed and to which we do not have, from you, the beginning of a solution.

That's completely inaccurate. We might need to start another thread to keep track of that particular (very interesting) sub-part of this discussion, but quite a few solutions, or at least solutions schemas, have been proposed for several of the known issues.

Quoting Mrocktor (Reply 228): Quoting Pihero (Reply 224): I do not think that you can guarantee that you've accounted for every possible failure on an aircraft. It is a requirement to demonstrate that you have accounted for every foreseeable failure.

Yes, but there's a big difference between foreseeable failure, possible failure, and unrecognized failure.

Quoting Rheinwaldner (Reply 229): Quoting Tdscanuck (Reply 223): If that's actually the best solution, why wouldn't you just program that in? Good lock with the use-cases of that program! And good luck with the acceptance testing of that feature! You have thought about testing of system features? If it is a feature it MUST be tested! How would a test series look like? Ten landing attempts each with a front gear that is arrested at different angles by 10 degree steps?

Yes. That's how you test a bunch of other airplane systems today (Vmcg, autoland, landing distance, etc., etc.). It wouldn't be certifiable if you couldn't demonstrate it safe, so the testing wouldn't be any riskier than what's done today.

Quoting Rheinwaldner (Reply 229): Quoting Tdscanuck (Reply 223): Most of today's emergency landings are caused by equipment failures, but most (the *vast* majority) are non-safety-of-flight failures, so why would they end in crashes? An engine failure often results in an emergency landing, but an engine failure is only considered an economic (not safety) event. Ok, you are right, the majority could be handled by an automated system. But the rest in absolute numbers surely still exceeds the number of crashes by pilot error.

I'm not at all sure about that. For quite a while, pilot error was the primary cause of crashes, so we have a pretty good idea of the size of that number. Emergency landings that are actually safety-of-flight failures are also pretty well known, so we know that number. I don't know either number exactly but, gut feeling, I suspect they're comparable.

Quoting Rheinwaldner (Reply 229): Quoting Tdscanuck (Reply 226): It's only failures that weren't anticipated that will potentially cause meaningfully divergent responses between a human and a computer. How could that gap be ever closed? It would require to anticipate every condition, also those that have never occurred before.

Actually, it doesn't. It required that you antipate every failure *symptom*, not every failure. This is still a big issue, but a much more tractable one that anticipating every failure, which is essentailly impossible. Just a trivial example, you don't need to anticipate all the ways that actuation of an aileron could fail. You just need to provide control authority sufficient to overcome an aileron in any position...then the types of failures that cause that symptom don't matter, since you know you can deal with all of them, even ones you didn't expect.

Quoting Rheinwaldner (Reply 229): IMO full scale testing of automated airliners would result in three to five hull losses per aircraft type. Otherwise you won't get that aircraft tested.

This could never happen. If you expected to lose 3-5 airframes, the aircraft is so unsafe that's it's certifiable and there'd be no point in testing in the first place, hence no loses. You'd work those level of glitches out with simulators, bench, and small scale (probably modified UAV) testing long before you got to full scale airliners.

Forgetting technology for a second, any business plan that involved cratering 3-5 airframes just wouldn't get off the ground (pun intended).

Quoting Rheinwaldner (Reply 231): The whole day at my workplace I struggle with computers that do one thing: cope with unexpected situations, just ... they fail miserably. Computers always stumble over unexpected situations. Computers never execute one command that has not been programmed.

They don't always stumble over unexpected situations, but the ones we work with on a day-to-day basis often do. Ones that are engineered where it really matters, like FBW and EEC's (or implanted medical, to pull from outside aviation) are considerably more robust. Also, although a computer can't execute a command that hasn't been programmed, that's actually not much of a limitation. Things like adaptive code and robust control can execute commands that result in outputs that were never explicitely coded.

Quoting Rheinwaldner (Reply 231): "One of the key concepts behind DO-178B is that all the requirements are covered by the code and that all the code that exists is directly due to a requirement." This implicates that any line of code is coupled to a requirement.

Yes.

Quoting Rheinwaldner (Reply 231): It categorically excludes anything like not delibertaly programmed responses.

No. Or, at least, not in the sense you mean. Adaptive control being the obvious aerospace example...the "deliberately programmed response" may be something like "keep the wings level." There may be a *huge* number of actuator solutions, any one of which can be executed at any time depending on damage, system function, atitude, etc. that will satify that programmed response.

Quoting Max Q (Reply 235): So we have a small group of people who, exclusively have no Pilot qualifications or experience in any real life use of their beloved computers in the cockpit or making the thousands of decisions covering myriad scenarios 24 / 7 / 365 days a year proposing all of these can be addressed autonomously.

Your issues with Bond007 notwithstanding, where did you get the outrageous (and manifestly false) assumption that the people on this thread who don't agree with you have no pilot qualifications or experience?

Quoting Max Q (Reply 235): When ridiculous ideas like letting Flight Attendants make decisions on whether or not to divert in life threatening situations are actually put forward as serious solutions most of us realise how much this group has totally lost touch with reality.

Are you seriously suggesting that an FA can't be trained/experienced to make a diversion decision? Obviously, a pilot can....what makes FA's inherently dumber than pilots?

Tom.

On the contrary I think Flight Attendants can be trained and experienced enough to make divert decisions and others.

Just send them to flight school to get all their ratings, than have them accumulate the several thousands of hours experience necessary working for smaller operators gaining the accumulated knowledge required while passing all the required academic, proficiency and medical tests on a regular basis.

After several years of vital experience they could start off in the right seat with a major airline after going through the selection process and passing numerous more academic and proficiency tests.

With a few more years of seasoning and the right level of competence then they could be considered for command and would eventually move to the left seat providing they passed all the required tests.

Then, my family and I would happily sit behind where they are seated in the cockpit, but of course wouldn't they then be called 'Pilots'

Quoting WPIAeroGuy (Reply 238): Everything now is designed for humans to operate. Thats why there are lights and painted lines on an airport. It would much easier to use GPS and perhaps and embedded EM path for computer navigation, because currently computers can do that better than humans.

Do away with the painting and the lighting ?
Are you suggesting that robot operations will be un-monitored ?
Or are you suggesting that robot monitoring will be done by more robots, which in turn will need more monitoring system...until the final stage where the human monitor will use lightings and painted boards to have an idea where the robot is...but it can be trusted so we'll replace him with a more fiable autonomous system...which of course will need monitoring....
Well, I really see the cleverness of that argument.

Quoting WPIAeroGuy (Reply 238): Oh I found the article. It was in Kitplanes which is why I couldn't find it. The article can be found here for subscribers

Just give us place/date/aircraft type/ and my search robot will find the event...if it ever happened in the real world, of course.

Quoting Tdscanuck (Reply 239): Hmm...maybe what I'm thinking of as dispatcher and what you're thinking of are different things.

No, not really. Just a matter of shared decision-making and split responsibilities. The bulk of the work is still using as fully as possible all the data available to ensure a smooth operation.
I see that *knowable* is more accurate...It doesn't change the fact that it implies data treatment which is probably the task computers are the best at. So, in my experience these data are as available to me as they are to the dispatcher.

Quoting Tdscanuck (Reply 239): It is a requirement to demonstrate that you have accounted for every foreseeable failure. Yes, but there's a big difference between foreseeable failure, possible failure, and unrecognized failure.

     

Quoting Tdscanuck (Reply 239): For quite a while, pilot error was the primary cause of crashes, so we have a pretty good idea of the size of that number.

We've already been through that argument. It took us - from the Wright brothers - some forty years to begin to recognise that flight deck environment needed to be better designed for safety : ergonomic, instrument presentation (the T), instrument readability and interpretability (cf the infamous "three-pointer-altimeter"which was at the origin of countless CFIT events...to the 737...) and then, we'll start qualifying "pilot error".
look, as you have realised, I am not against automated systems and having flown DC-4s and Airbuses, I certainly can vouch for the increased safety that a well-thought human-machine interface can bring, and moreover, through the stages of experience on ND26, 737, 1011, 741/2/3/4 I can see how much progress has been achieved. The tandem pilot/system works well but we still find areas where the design has erred for lack of foresight and potential accidents could still be happening.
And I really do not see why one would think of removing the pilot from the airplane, to have someone do what he does the best -i.e appraisal and decision making- through an array of data link / crosstalking gizmoes and whatnots. To me that's the most ludicrous idea behind the concept.
So, if I understand well, everybody can be trained as part of the future systems : dispatchers...engineers/joystick operators when the chips are down....stewardesses to make decisions on-board their automated planes... lawyers to fight new concepts of responsibility in courts...who else ?...but the pilots have to go ?
Right ?

Quoting WPIAeroGuy (Reply 232): If a piloted airliner doesn't have to be tested for it, why would a pilotless one?

Features of a system have to be tested. That is true especially for such critical applications. Many of the required systems have to do with the physics. If the control path of a controller involves physics then physics have to be tested. The proposed automated aircraft has so many (new) physical control paths, that a much higher effort for severe testing with real machines would be required.
Therefore I said that testing of all those systems could easily cause several hull losses per type.

If you see the reliability requirement (robustness) in addition (those systems are required to have a stellar MTBF) the cost will rocket up. In history much easier tasks (easier to automate) have never been automated e.g. the tasks of an airline boss. Why do not all railways make use of automated trains? There exist some but most (all?) new ones are not. Though an automated train is so much easier to accomplish than an automated plane.

Quoting WPIAeroGuy (Reply 238): Flight attendants will need additional training to become a better interface for what happens in the cabin and what the flight computer decides to do.

Automation usually substitutes many simple workers by a few highly skilled operators. That you promote the opposite is remarkable and may indicate some flaws with the intention.

Quoting Tdscanuck (Reply 239): so the testing wouldn't be any riskier than what's done today

Much riskier and comprehensive (caused by multiple complexity). Based on present flight test durations I expect an new automated airliner to be in flight test for several years.

Quoting Tdscanuck (Reply 239): No. Or, at least, not in the sense you mean. Adaptive control being the obvious aerospace example...the "deliberately programmed response" may be something like "keep the wings level." There may be a *huge* number of actuator solutions, any one of which can be executed at any time depending on damage, system function, atitude, etc. that will satify that programmed response.

Modern controller theory is able to deliver amazing results. Still the base are mathematic calculations and those formulas can be validated in theory and labs. "Keep wings at level" is expressed by a mathematic formula. The software only implements that formula. The mathematic correctness of the controller can be tested and the outcome on the real machine is predictable.
But the requirements for an automated-plane-guidance-system demand for much more improved systems. To substitute pilots those systems should be able to develop initiative in an absolutely unacceptable manner. IMO here lays an unreachable target.

Quoting Tdscanuck (Reply 239): Are you seriously suggesting that an FA can't be trained/experienced to make a diversion decision? Obviously, a pilot can....what makes FA's inherently dumber than pilots?

Less training, means less costs, but when F/A's get more costly the business case for pilotless planes diminishes. I mean if you replace pilots by "pilots" who serve passengers you don't gain much.

Quoting Pihero (Reply 241): So, if I understand well, everybody can be trained as part of the future systems : dispatchers...engineers/joystick operators when the chips are down....stewardesses to make decisions on-board their automated planes... lawyers to fight new concepts of responsibility in courts...who else ?...but the pilots have to go ? Right ?

  

Quoting WPIAeroGuy (Reply 232): Another idea we've skipped over is the networking of these aircraft. Each one could learn, in real time, from every other airplane around the globe. For instance, based on the weather radar, airplane X decides to take a certain route. Accelerometers measure turbulence, and determine the ride is too rough for passengers, and a new route is determined. This decision is then broadcast to all other autonomous aircraft, so when faced with similar situations they're decision making is slightly altered.

Very promising on the feature side but adding tons of more complexity. Issues with "going offline" are unsolvable. Either that feature is not really required for save operation (like thrust reverses) or networking reliability issues kill the application. And ... if its not required for save operation why build it?

Quoting Pihero (Reply 233): In most of these cases the designers, manufacturers and maintenance personnel "failed to anticipate..." or "did not consider..." some combination of factors or occurrences.

Quoting Pihero (Reply 233): While supporting such access is often rejected based on statistics implicating the pilot as the top causal factor in current accidents, it is extremely important to recognize that no comparable data is recorded on successful human interventions (indeed, we are not sure how such data could be collected), that is, the number of failures, errors, and ultimately accidents that are avoided or remedied by human actions. Without data on beneficial interventions, using accident statistics to advance the case that human pilots "usually" make things worse and thus their system access should be minimized or eliminated is misleading and would be like saying that because engine failures lead to crashes, engines cause aircraft to crash and should be eliminated.

That is clearly expressed what I tried to explain   

Quoting Tdscanuck (Reply 239): For quite a while, pilot error was the primary cause of crashes, so we have a pretty good idea of the size of that number. Emergency landings that are actually safety-of-flight failures are also pretty well known, so we know that number. I don't know either number exactly but, gut feeling, I suspect they're comparable.

It is not only emergency landings. There are a lot of ambiguous situations that pilots could easily assess to go on. Automated system that are designed to the "safe side" produce always a lot of false alarms. This would result in unnecessary diversions, collapsing dispatch rates because computers measuring wrong (even if the system checks are bullet proof the dispatch rates probably collapses because the much complexer systems generate multitudes of errors anyway). You know, pilots have a pretty good dispatch rate! Their task can easily and in no time be substituted by stand-by pilots.

Quoting Rheinwaldner (Reply 231): Aircraft automation must make use of software that is written after "DO-178B": This standard requires amongst many other things:

Yes, I am a member of RTCA (although I'm lying, so take it with a pinch of salt) and I'm very familiar with DO-178B and many other 'guidance' standards related to avionics and software. In fact one of the shortcomings of DO-178B is it's flexbility ... well, and the fact it's a guidance document.

But regardless, you are doing the same as others ... comparing new technology and concepts with current standards. RTCA are creating new standards and documents every year ... in fact I attend the discussions on a couple of them. Let's make an educated assumption that standards and regulations will have to change, shall we. Let's face it ... 95% of Part 121 is irrelevant if/when we have automated aircraft.

Quoting Rheinwaldner (Reply 231): The whole day at my workplace I struggle with computers that do one thing: cope with unexpected situations, just ... they fail miserably.

Is that Windows Vista perhaps? A Mac?
Yes, my PC just had to reboot ....
We won't be running aircraft with $499 desktops, or $99 software.

Quoting Rheinwaldner (Reply 231): Computers never execute one command that has not been programmed.

Perhaps, but again .... not what I said.

Quoting Rheinwaldner (Reply 231): Quoting Bond007 (Reply 230): The major misunderstanding here, is the thinking that without a list of all the millions of possible scenarios somehow programmed into the computer, it cannot cope with unexpected situations.

You do not have to program each and every specific scenario. A simple example is recovery from unusual attitudes. Does a pilot really have a checklist, mentally or physically, of the 16 (or more) possible specific combinations of attitudes, or does he simply know 3 or 4 control responses 'programmed' for individual subsets (climbing, banking, stalling etc.) ? They tell me that they only teach you how to recover from perhaps a couple of examples of these, but you are somehow instantly capable of recovering from many times more scenarios than you were 'programmed' for ... you don't need to be taught every one, that's why ... just the responses for specific inputs from instruments.

Quoting Max Q (Reply 235): So we have a small group of people who, exclusively have no Pilot qualifications or experience in any real life use of their beloved computers in the cockpit or making the thousands of decisions covering myriad scenarios 24 / 7 / 365 days a year proposing all of these can be addressed autonomously.

You're doing it again, and it's a prime example of somebody who has little to add to his/her argument, so resorts YET AGAIN to attacking the people making the arguments, instead of simply attacking the argument itself .... it's text book debating stuff here, and quite tiresome. You had a whole post that contained little but ad hominen attacks!

...as for the F/A discussion, why don't you all read the original scenario. I'm truly bored of repeating it. We are discussing having something/somebody that can determine the different between burnt coffee and burnt wiring insulation ... and the F/A made that decision in the scenario described, and passed the info to the cockpit. The same can happen with a F/A in the future, or numerous types of gas/smoke sensors, many of which can distinguish a fart from a burning airplane.


Jimbo

Quoting Bond007 (Reply 243): Is that Windows Vista perhaps? A Mac? Yes, my PC just had to reboot .... We won't be running aircraft with $499 desktops, or $99 software.

PC, huge business application, caused a tremendous effort, 1/4 million lines of code, development of successor is estimated to cost 20 millions in cash. A lot depends on it that it runs but no lives depend on it.
I can't imagine in my wildest dream to make a software of that complexity reliable enough to make it rule over lives. I consider software development as a highly unexplored terrain. The answers that want to get a grip over the outcome of software projects seem so helpless. If I apply all methods that are promoted by the theory (requirements management, configuration management, test management, deployment management) I get so consumed by management duties that my output in terms of lines-of-code tends towards zero and my role becomes a MS office pirate!

Quoting Bond007 (Reply 243): They tell me that they only teach you how to recover from perhaps a couple of examples of these, but you are somehow instantly capable of recovering from many times more scenarios than you were 'programmed' for ... you don't need to be taught every one, that's why ... just the responses for specific inputs from instruments.

As posted before there is controller like functionality (limited and clearly specified in terms of input parameters) that can be realized with computers (in fact most controllers can be realized with analog electronics equally well). Basically they do implement mathematic formulas. Here is no problem. To automate a complex desicioning system (but still much more primitive compared to tasks executed by a pilot) you would implement something like a rules-engine. Also a rule engine develops no initiative. From that point to go to a realy "creative" software there is another huge gap. And that gap can not be overcome on the way to an automated plane. "creative" software on a plane? Please never!

Quoting Rheinwaldner (Reply 244): PC, huge business application, caused a tremendous effort, 1/4 million lines of code, development of successor is estimated to cost 20 millions in cash. A lot depends on it that it runs but no lives depend on it.

Do you trust an AED? There might not be as many lines of code and its not as complex, but it still must interperet electrical impulses from the heart to determine if there is fibrillation, and then determine the correct shock to apply. People depend on computers to save their lives every day.

Quoting Rheinwaldner (Reply 242): Quoting WPIAeroGuy (Reply 232): If a piloted airliner doesn't have to be tested for it, why would a pilotless one? Features of a system have to be tested. That is true especially for such critical applications. Many of the required systems have to do with the physics. If the control path of a controller involves physics then physics have to be tested. The proposed automated aircraft has so many (new) physical control paths, that a much higher effort for severe testing with real machines would be required. Therefore I said that testing of all those systems could easily cause several hull losses per type.

I don't think I was clear, I didn't mean that the software wouldn't be tested, I meant that the flight tests wouldn't be much more complex than what is performed today. The systems testing would be expansive, true. I'm pretty sure part of the certifcation process doesn't involve a gear-up landing, but I know the airplane must be capable of one. Why would a computerized plane have to perform an actual gear up landing, as long as the same mechanical analysis that has always been done is completed, and the software performs correctly in the simulator?

WPI,

You know that's a bad example (The AED). People fly planes WAAAAY more often than they use an AED. Also planes operate for a far longer duration.


Blackbird

Quoting Rheinwaldner (Reply 244): I can't imagine in my wildest dream to make a software of that complexity reliable enough to make it rule over lives.

Luckly the rest of humanity is not constrained by your imagination. Or lack thereof. Any modern FBW aircraft has millions of lines of code (as in more than one). Feel free not to fly the A380, A330/340, 777 or E-jets if you can't trust our imagination.

If anyone has the book handy, I can't recall the quote nor do I have the book anymore, but in the first chapter of Dale Brown's Dreamland, he sums up this entire thread in one line that goes something like, 'The precision of computers will never replace human creativity and ingenuity.' I'll edit this post if I can find the quote, I'm going to keep searching online ...

Quoting Blackbird (Reply 246): You know that's a bad example (The AED). People fly planes WAAAAY more often than they use an AED. Also planes operate for a far longer duration.

The example was in reference to people trusting their lives to automation. It has nothing to do with how often its used, it has to do with the fact that when it is used, there are no experts present and therefore the lifesaving decision is up to the computer.