Single-pilot airliners and AI (split from "Envoy captain stricken after takeoff and passes")

[nauga]

Two ends arguing about the middle.

In my experience, much of which is applicable to this thread, line pilots often are not aware of the state of the art and rate of progress in guidance, navigation, and control; and the developers of this technology are often not aware of the details of certification in a flight critical application. What is possible is not necessarily what is certifiable, and until the FAA accepts control laws that may not be deterministic or may not be suited to commonly accepted means of validation and verification the technology will be confined to applications that are more willing to accept the risk (read: military). Full authority FBW was well matured by the time civil applications became practical, and people still complain about pilots not having direct yoke-to-surface control of airplanes, but it is in widespread use. OTOH, it is naive to fail to recognize that customers (read: carriers) are funding development of autonomy in passenger-carrying applications - it may be a long-shot and far horizon technology but there is interest and progress.

Nauga,
who doesn't think there is the equivalent of certification DAL in automotive applications

 

[nauga]

I don’t have any direct knowledge but I believe there is already an intense race underway to produce a pilotless military aircraft - whoever gets there first will have the advantage of being able to utilize flying contraptions that can routinely pull 12+ Gs plus host of other benefits.

A lot of what's being discussed here, including this, does not require AI.

Nauga,
Fox 2

 

[Bell206]

Outside of a FAR no one really cares what the jokers at the FAA think

FYI: its already an FAR. The guidance to complying with that FAR is what changed and this AC is one of the acceptable options just released. But I'm sure you knew that. Regardless, manual flight skills and automation has been pointed discussion over the last 6 years resulting in congressional mandates and the like which is basically the opposite direction of AI/single pilot ops.

 

[nauga]

Regardless, manual flight skills and automation has been pointed discussion over the last 6 years resulting in congressional mandates and the like which is basically the opposite direction of AI/single pilot ops.

I'll be interested to see the mandate to turn off all stability augmentation and conduct normal ops in whatever flavor of direct/manual/MECH/DEL 'your' 21st century airplane has. Turning off all 'automation' is very different from turning off pilot relief modes like autopilots.

Nauga,
a student of the arts

 

[TCABM]

Technology, ......ology! AI? Phoey! Let's talk today and real world.
Has anybody had an actual incident of inflight incapability? Thats where this thread should go. I'll toss in an actual experience to begin.
Part 135, two pilot crew. My co-pilot passed a gall stone while over water and one hour from land. He had a recent Cl 2 medical.
Next?

Ok. The guy in the article below was a co-worker when this happened. What would you like to know?

He’s also now employed by that airline

https://www.cnn.com/2014/06/03/travel/737-emergency-pilot

 

[nauga]

What would you like to know?

Why are we still relying on meat servos? They're non-deterministic, have latent unpredictable failure modes, and are the root of the leading causal factors of mishaps.

I'm being intentionally facetious, but also pointing out that there are two ways of looking at this problem. The smart technologist hedges his bets.

Nauga,
meatier than some

 

[TravisRNT]

FYI: its already an FAR. The guidance to complying with that FAR is what changed and this AC is one of the acceptable options just released. But I'm sure you knew that. Regardless, manual flight skills and automation has been pointed discussion over the last 6 years resulting in congressional mandates and the like which is basically the opposite direction of AI/single pilot ops.

We already removed spin training, the complex requirement for a CPL, even FSI these days is teaching more and more rote


I can take many a fellow working pilot up in a plane, get it to stall and drop a wing and 9 out of 10 first instinct is to lift the wing with roll vs stepping on the high wing and unloading the plane.

I don’t think adding pounds to the FAR or relying even more on automation is going to fix that

Why people think the same government who makes these problems will somehow be the solution to all their problems, we’ll let’s just say I’ve had my name on the list for the mars expedition for a long time lol

 

[Bell206]

I'll be interested to see the mandate to turn off all stability augmentation

Doubtful they'd go to the SAS levels but the current law states it this way:

I don’t think adding pounds to the FAR or relying even more on automation is going to fix that

While automation has definitely improved safety, in the past 15 years or so there's been an inverse move to it: loss or misunderstanding of the automation causing accidents. In some circles addressing the issue has been equated to putting the "aviate, navigate, communicate" back into the modern cockpit. Time will tell.

 

[TravisRNT]

Doubtful they'd go to the SAS levels but the current law states it this way:
View attachment 112586


While automation has definitely improved safety, in the past 15 years or so there's been an inverse move to it: loss or misunderstanding of the automation causing accidents. In some circles addressing the issue has been equated to putting the "aviate, navigate, communicate" back into the modern cockpit. Time will tell.

Automation is simply a work load reduction tool, feels like in recent years, especially in the 3rd world, it’s being used as a pilot skill reduction tool, time will tell however I don’t think you need a magic 8 ball to guess what it will tell

Watching the flight channel on YouTube, highly recommend, many accidents on there where instead of kicking the AP YD off and feeling the plane to see what they are working with, the crew defaults to keeping it on until things are beyond recoverable and even trying to turn it back on when it kicks off due to major engine issues or flight control issues

 

[nauga]

Doubtful they'd go to the SAS levels but the current law states it this way:

Yeah, I don't disagree with the law, which specifically states 'autopilot or autoflight,' my issue is with those who condemn 'automation' as a whole without fully understanding what's automatic and what's not.

Case in point:

Automation is simply a work load reduction tool...

It *is* a workload reduction tool, but in some cases that workload reduction can make an otherwise unflyable airplane flyable. Turning *that* off to fly the unaugmented airplane is not recommended, regardless of your skill level.

Nauga,
who is not mechanical but is reversible

 

[falconkidding]

Watching the autopilot and throttles struggle with an ILS in windy conditions always makes me feel better about automation taking my job. I think cerification will be the hangup. Ensuring that it could handle unexpected situations is gonna take so much testing. This year i had a terrain warning on a rnav in visual conditions. No biggie visual so continue, turns out was a jepp database error. How would you certify an AI to ignore a terrain warning? I could see the path to how you would teach an AI that but to certify that would be insane. Get jammed into LGA s turns approved. Hows the certification work on s turning on final to 500 ft seeing where the other guy touched down and making the judgement call to continue. I can see how an AI could fly that but how do you certify an autonomous aircraft for "s turns for spacing"

 

[TCABM]

Why are we still relying on meat servos?

We design for the technology we have, not the technology we want.

 

[nauga]

We design for the technology we have, not the technology we want.

If you wait for the technology you need to be mature to start your design, you're behind before you start.

ETA: In aeropace, at least, there are very few serious technology houses advancing tech for the sake of tech. They are advancing it because there's a need for that tech.

Nauga,
who pushes and pulls

 

[TCABM]

The biggest challenge I see isn’t the automation part, it’s training the Artificial Intelligence model and then getting that model certified, than integrating it into the various systems and then getting the airframe certified.

 

[TravisRNT]

The biggest challenge I see isn’t the automation part, it’s training the Artificial Intelligence model and then getting that model certified, than integrating it into the various systems and then getting the airframe certified.

The non AI systems have enough logic issues as it is, talking about AI at this point is like talking to a 600lb man about losing weight after you pull a wall out to get him out of his home, only to have him talk about needing a coach to help him win the upcoming Olympics track and field events

Got new airplanes with auto throttles and APs that still hunt and do the magic snake dance when intercepting, and we are talking about AI?

The no pilot airliner engineer/science types are just like the people who tell me the temperature my state is going to be in 50 years, but can’t give even give me good forecast 48hrs out

Baby steps

 

[Bell206]

The biggest challenge I see isn’t the automation part

I think it actually includes all the above especially when dealing with advanced designs like AI. They still have problems integrating developed, mature technologies/designs into "new" platforms. Just look at the Bell 525 and the AW609 projects. Both are the 1st FBW commercial rotorcraft and powered lift aircraft going through certification. Unfortunately each suffered catastrophic flight test failures due to control law anomalies that were unforeseen in any simulations or previous tests. Regardless, innovation still needs to continue but for those people who are publicly pushing for single pilot/AI airlines now are following an agenda more than a reality.

 

[Maxnr]

I think it actually includes all the above especially when dealing with advanced designs like AI. They still have problems integrating developed, mature technologies/designs into "new" platforms. Just look at the Bell 525 and the AW609 projects. Both are the 1st FBW commercial rotorcraft and powered lift aircraft going through certification. Unfortunately each suffered catastrophic flight test failures due to control law anomalies that were unforeseen in any simulations or previous tests. Regardless, innovation still needs to continue but for those people who are publicly pushing for single pilot/AI airlines now are following an agenda more than a reality.

A little update on the Leonardo/AW609 crash a few years ago. The test AC was flight testing an aerodynamic clean up at about 300 KTS. The BAE Tripple redundant auto pilot was engaged. A slight left aileron was inappropriately corrected with a tad too much rudder. The tinkering with the AP programing was not confirmed in the sim. The yaw couple started a dutch roll which was inappropriately dealt with by the test pilot who countered it with roll corrections. The roll increased in amplitude until the AC broke up. The FDR showed that the pilot had applied full deflection roll inputs. Anybody that has flown a Sikorsky, with AP off, knows that dutch roll is countered with yaw inputs.
The Bell 214ST was designed for the Shah of Iran around 1972 and had a fly-by-wire sync elevator that is turbulence sensitive. One actuator will go up and the other down. Hated it.

 

[Sundancer]

Probably closer for aircraft than cars - fully autonomous and ubiquitous Level 5 cars may be a long way in the future. As in, the vehicle pulls up out front, you get in, and it drives you to to work (or to Miami from Denver) while you nap or watch YouTube in the backseat.

Work on it has been going on for decades - I think the direct connection to today's efforts are from the 70's. Other experiments go back much further, with wires imbedded in roads for guidance, lights on the shoulders,etc. Lot of work to be done on sensors, AI, etc.

 

[nauga]

Got new airplanes with auto throttles and APs that still hunt and do the magic snake dance when intercepting, and we are talking about AI?

Autothrottles and autpilots are not new technology. I suggest that issues like your example are not related to the technology itself but the implementation and in some cases budget. There are also new airplanes with no control augmentation that have miserable flying qualities in some parts of their flight envelope, but no one is suggesting that cable and pushrod technology is not sufficiently mature.

Nauga,
and Cooper and Harper

 

[Warmi]

Th

Probably closer for aircraft than cars - fully autonomous and ubiquitous Level 5 cars may be a long way in the future. As in, the vehicle pulls up out front, you get in, and it drives you to to work (or to Miami from Denver) while you nap or watch YouTube in the backseat.

Work on it has been going on for decades - I think the direct connection to today's efforts are from the 70's. Other experiments go back much further, with wires imbedded in roads for guidance, lights on the shoulders,etc. Lot of work to be done on sensors, AI, etc.

Neural networks are not new and are in fact decades old concept - what really makes the difference is available computing power/storage/connectivity.
Your iPhone is something around million times more powerful than the the original NASA computers from the 1960s and 70s… and that just your tiny personal IPhone. Now factor in large arrays of graphics processors , each many times more powerful than your IPhone … all of that power networked and working in unison crunching numbers… Skynet is coming for us

 

[Piperonca]

Autothrottles and autpilots are not new technology. I suggest that issues like your example are not related to the technology itself but the implementation and in some cases budget. There are also new airplanes with no control augmentation that have miserable flying qualities in some parts of their flight envelope, but no one is suggesting that cable and pushrod technology is not sufficiently mature.

Nauga,
and Cooper and Harper

Mr. Cooper got into wine, too...

https://garrodfarms.com/test-pilot-blends/

 

[nauga]

Mr. Cooper got into wine, too...

https://garrodfarms.com/test-pilot-blends/

I didn't know that! No one ever asked me to rate wine, but if asked I would probably have given everything a 10 on Cooper & Harper's scale - "Control will be lost during some portion of the required operation."

Nauga,
Chasing the elusive 4.5

 

[FPK1]

HAL, land the plane. … Siri, why isn't HAL responding? … Alexa? … Cortana? … Anyone?

Sent from my Pixel 7 Pro using Tapatalk

 

[Piperonca]

I didn't know that! No one ever asked me to rate wine, but if asked I would probably have given everything a 10 on Cooper & Harper's scale - "Control will be lost during some portion of the required operation."

Nauga,
Chasing the elusive 4.5

Learned about it when our well-respected Aeromechanics Division Chief retired a few years ago, with every honor in the book. His old friends from what had been the Army Edwards Flight Activity (or maybe it was the group at Moffett Field, I don't know), sent a bottle of Test Pilot wine for his ceremony. I'm pretty sure it was purchased and hand-carried by the Senior Research Scientist from California to Alabama. Too bad the attendees didn't get to rate it.

 

[nauga]

HAL, land the plane. … Siri, why isn't HAL responding? … Alexa? … Cortana? … Anyone?

Sent from my Pixel 7 Pro using Tapatalk

Hey Google, stop automatically attaching adverts to the end of all my posts.

Nauga,
who doesn't distrust jet engines just because his toaster burns his pop-tarts

 

[TravisRNT]

Hey Google, stop automatically attaching adverts to the end of all my posts.

Nauga,
who doesn't distrust jet engines just because his toaster burns his pop-tarts

If the toaster is burning the tarts, lay off the TRs

 

[TravisRNT]

Autothrottles and autpilots are not new technology. I suggest that issues like your example are not related to the technology itself but the implementation and in some cases budget. There are also new airplanes with no control augmentation that have miserable flying qualities in some parts of their flight envelope, but no one is suggesting that cable and pushrod technology is not sufficiently mature.

Nauga,
and Cooper and Harper

I’m taking new system in new aircraft

 

[nauga]

I’m taking new system in new aircraft

Maybe you're talking about a new implementation in a new aircraft but as before, civil autothrottles and autopilots are not new technology and are well understood. New implementation issues with these systems are typically not related to the technology *of those systems*. If issues with them make it through testing and into the fleet it's not due to some flaw or shortcoming inherent in the technology.

Nauga,
fast, cheap, or good

 

[TravisRNT]

Maybe you're talking about a new implementation in a new aircraft but as before, civil autothrottles and autopilots are not new technology and are well understood. New implementation issues with these systems are typically not related to the technology *of those systems*. If issues with them make it through testing and into the fleet it's not due to some flaw or shortcoming inherent in the technology.

Nauga,
fast, cheap, or good

So if our current abilities can’t get a simple auto throttle and AP to work decently, why even talk about AI?

 

[Sluggo63]

Maybe you're talking about a new implementation in a new aircraft but as before, civil autothrottles and autopilots are not new technology and are well understood. New implementation issues with these systems are typically not related to the technology *of those systems*. If issues with them make it through testing and into the fleet it's not due to some flaw or shortcoming inherent in the technology.

Nauga,
fast, cheap, or good

Usually, I would concur. But sometimes they aren't well understood. Here's a Safety Bulletin from our current flight publications, detailing an autothrottle problem that has existed since at least 2016. They still can't figure out the problem.


So, basically, the computer on its own decides that, while on the ground taxiing around, "now would be a great time to go to TOGA thrust."

Boeing's solution? Have the meat servo in the left seat keep their hands on the thrust levers to prevent the aircraft from going full thrust while number 13 in line at JFK waiting to takeoff. Easy solution? Sure. The most elegant? Nope. But (as you know) although the human element in the design is the weakest link, there are other times when the answer is "just have the captain hold the thrust levers in idle."

Sluggo 63
and 200 pounds of Grade A Prime

 

[Jim K]

I think the certification of garmin's autoland opened pandora's box. I think the first thing we'll see is single pilot jets with a deadman switch in the form of a camera monitoring their eye movements or some such. In the event of pilot incapacitation, the plane declares an emergency and lands itself. No dealing with weird atc instructions or worries about a reliable data link to a ground based pilot.... just get out of my way I'm coming in.

 

[nauga]

So if our current abilities can’t get a simple auto throttle and AP to work decently, why even talk about AI?

Our current technology is more than capable of getting a simple autothrottle and AP to work decently and has been so for decades. That doesn't mean every implementation is flawless. Same holds for pilots and pilot training, yet we still talk about training more pilots.

Nauga,
who is not as anti-pilot as he seems

 

[Greg Bockelman]

But (as you know) although the human element in the design is the weakest link, there are other times when the answer is "just have the captain hold the thrust levers in idle."

Wouldn’t a better solution be to put “Auto Throttles—Armed” on the Final Items part of the Before Takeoff checklist? Or does the problem show up even if the autothrottles are off!

 

[TCABM]

… That doesn't mean every implementation is flawless.…
Nauga,
who is not as anti-pilot as he seems

That’s kind of the point of my comment about designing for the technology we have…even the best AI implementations will likely require man-in-the-loop interrupt capability, precisely because there will be instances of undesired/undesirable actions occurring.

The meat servo is the last chance, and by that logic, where system redundancy is required, a second meat servo will be required because the AI model can’t be interviewed by an NTSB investigator.

 

[Warmi]

That’s kind of the point of my comment about designing for the technology we have…even the best AI implementations will likely require man-in-the-loop interrupt capability, precisely because there will be instances of undesired/undesirable actions occurring.

The meat servo is the last chance, and by that logic, where system redundancy is required, a send meat servo will be required because the AI model can’t be interviewed by an NTSB investigator.

That’s probably most likely route - full automation with a single pilot watching it. This setup will still require periodic manual trips to keep pilots current ( or alternatively rely on 90% Flight Sim training which will keep getting better and more realistic ) but even that will save airlines millions in personnel cost and make whole industry even safer.
I do believe incredible advances in safety we been enjoying in the last 10-15 years are mostly due to technology ( all sorts - GPS, automation etc )

 

[MauleSkinner]

That’s probably most likely route - full automation with a single pilot watching it. This setup will still require periodic manual trips to keep pilots current ( or alternatively rely on 90% Flight Sim training which will keep getting better and more realistic )…

At that point you’re probably just as well off to leave the pilot out of it completely…people grossly underestimate the value of flying airplanes, conversations about “what it” that pilots often have, or simply the value of a second set of eyes to confirm that yes, in fact, this isn’t a normal indication.

 

[Palmpilot]

I'm trying to figure out how to work "meat servo" into my sig line.

 

[MauleSkinner]

I'm trying to figure out how to work "meat servo" into my sig line.

“Self deception regarding one’s meat servo is the most dangerous kind.”?

 

[Palmpilot]

“Self deception regarding one’s meat servo is the most dangerous kind.”?

Good suggestion, but I suspect that the meat CPU is also a risk factor!

 

[nauga]

I'm trying to figure out how to work "meat servo" into my sig line.

It works well as part of one of those LinkedIn trigrams like "Thinker. Dreamer. Maker."

Nauga,
Coffee Holder. Sentient Ballast. Meat Servo.