Two Pilot Long Haul Ops? Airbus & Cathay
#81
Gets Weekends Off
Joined APC: Mar 2020
Posts: 218
Two of the issues I have with the race towards automation. First is how often have we as pilots have to step in and correct dispatch over something. Not that it happens all the time. But in my 20+ year career, every so often I come across something in a flight plan that is not acceptable, either for safety or for passenger comfort. There is something that happens to how you look at and review a flight plan when you known you are going to have to live with it, and not just fire it off and sit in some cozy office sipping coffee. There have been a couple of times in my career where I flat out refused a flight plan and when I had to explain it to the dispatcher, their response was "oh...yeah...I guess you're right. I didn't think of that". Add to that, how many times have we seen a storm or bad weather that wasn't accurately displayed on there weather radar, or even did not appear at all? I just think you are going to lose a level of safety and comfort when you take pilots out of the equation, no matter how good the automation can get.
The second issue is protecting the data link between the ground and the airplane. I just flew a trip a few days ago with a guy who just got out of the Air Force right before the pandemic hit. His last assignment was flying the Predator (can't remember the model). We were just casually chatting about some of the things he did, and one common theme that came up was the accident rates. While he said most of the accidents were related to the fact that they are single engine, there have been a few that were due to loss of datalink. In fact there was a rather expensive loss just a few years ago that was simply due to the datalink being lost. These are highly secured datalinks, with connection speeds that are considered classified, and they still occasionally fail. On top of that, foreign governments are working hard at trying to figure out how to hack the links. Imagine what would happen if a datalink was lost on a passenger airliner? Imagine the devastation to the industry if North Korea announced it had figured out how to hack the datalink for a major U.S. airline?
I know right now all we are talking about here is a single pilot on the flight deck during cruise of a long haul flight, but this is obviously the direction they are trying to go.
Regarding single pilot ops, someone earlier mentioned the workload during a diversion. I will second this sentiment. I was in the bunk when we had a medical diversion in to Goose Bay. While there was no aircraft issue, the workload during the event was rather high for three of us. I can only imagine what it would be like during an aircraft emergency. On top of that, I was completely out when the call came to the bunk. Getting to the point of being fully awake and able to think clearly took about ten minutes or so.
The second issue is protecting the data link between the ground and the airplane. I just flew a trip a few days ago with a guy who just got out of the Air Force right before the pandemic hit. His last assignment was flying the Predator (can't remember the model). We were just casually chatting about some of the things he did, and one common theme that came up was the accident rates. While he said most of the accidents were related to the fact that they are single engine, there have been a few that were due to loss of datalink. In fact there was a rather expensive loss just a few years ago that was simply due to the datalink being lost. These are highly secured datalinks, with connection speeds that are considered classified, and they still occasionally fail. On top of that, foreign governments are working hard at trying to figure out how to hack the links. Imagine what would happen if a datalink was lost on a passenger airliner? Imagine the devastation to the industry if North Korea announced it had figured out how to hack the datalink for a major U.S. airline?
I know right now all we are talking about here is a single pilot on the flight deck during cruise of a long haul flight, but this is obviously the direction they are trying to go.
Regarding single pilot ops, someone earlier mentioned the workload during a diversion. I will second this sentiment. I was in the bunk when we had a medical diversion in to Goose Bay. While there was no aircraft issue, the workload during the event was rather high for three of us. I can only imagine what it would be like during an aircraft emergency. On top of that, I was completely out when the call came to the bunk. Getting to the point of being fully awake and able to think clearly took about ten minutes or so.
#82
I don’t think certified autonomous passenger aircraft are coming anytime soon… but self-driving cars are a red herring. Self driving cars are a vastly harder computational problem because roads are so non-standard let alone reacting to other drivers, pedestrians, animals and FOD. Airplane autonomy has a much narrower set of conditions to handle in normal ops and a vastly higher hardware budget per unit.
when airplanes fly themselves really isn’t related to when cars drive themselves.
when airplanes fly themselves really isn’t related to when cars drive themselves.
Aside from weather, the airplane's environment is more predictable, but you still do have to account for weather. Dumb AI can easily deal with that by strict avoidance but then you get a lot of diverts and cancellations... that's one of those things where human judgement is unmatched.
Also ground vehicles just have to not hit anything... that part's easy for airplanes and they can do it right now (ATC separation, TCAS, EGPWS). The big AI challenge for aircraft is systems failures and unanticipated UAS... a car can just pull off the road and auto-dial the tow-truck, but the airplane needs to keep flying, deal with the issue, and get on the ground safely.
It's the near-infinite possible permutations of failures, weather, geography, etc which make it really hard to program for... that's also the reason ground vehicle autonomy has stalled, they simply can't program for every possible collision avoidance permutation on the road. And they certainly don't have anything smart enough to figure it out for itself.
Remember, an autonomous airplane has to actually complete flights to be economical... it can't punt with a divert or cancellation every time it gets off the script.
#83
In a land of unicorns
Joined APC: Apr 2014
Position: Whale FO
Posts: 6,470
Yes and no.
Aside from weather, the airplane's environment is more predictable, but you still do have to account for weather. Dumb AI can easily deal with that by strict avoidance but then you get a lot of diverts and cancellations... that's one of those things where human judgement is unmatched.
Also ground vehicles just have to not hit anything... that part's easy for airplanes and they can do it right now (ATC separation, TCAS, EGPWS). The big AI challenge for aircraft is systems failures and unanticipated UAS... a car can just pull off the road and auto-dial the tow-truck, but the airplane needs to keep flying, deal with the issue, and get on the ground safely.
It's the near-infinite possible permutations of failures, weather, geography, etc which make it really hard to program for... that's also the reason ground vehicle autonomy has stalled, they simply can't program for every possible collision avoidance permutation on the road. And they certainly don't have anything smart enough to figure it out for itself.
Remember, an autonomous airplane has to actually complete flights to be economical... it can't punt with a divert or cancellation every time it gets off the script.
Aside from weather, the airplane's environment is more predictable, but you still do have to account for weather. Dumb AI can easily deal with that by strict avoidance but then you get a lot of diverts and cancellations... that's one of those things where human judgement is unmatched.
Also ground vehicles just have to not hit anything... that part's easy for airplanes and they can do it right now (ATC separation, TCAS, EGPWS). The big AI challenge for aircraft is systems failures and unanticipated UAS... a car can just pull off the road and auto-dial the tow-truck, but the airplane needs to keep flying, deal with the issue, and get on the ground safely.
It's the near-infinite possible permutations of failures, weather, geography, etc which make it really hard to program for... that's also the reason ground vehicle autonomy has stalled, they simply can't program for every possible collision avoidance permutation on the road. And they certainly don't have anything smart enough to figure it out for itself.
Remember, an autonomous airplane has to actually complete flights to be economical... it can't punt with a divert or cancellation every time it gets off the script.
20 to 30 years, maybe, for the technology to be available. Another 10-15 for the red tape.
Someone born today could still have a good career as an airline pilot.
People confuse machine learning and AI all the time. You can have bots that have seemingly meaningful conversations, and identify images easily. But they dont have any of I in the AI.
#84
The AI required for this level of decision making is still in the pipe dream stage. We have done huge leaps in machine learning, but any sort of credible AI just does not exist today.
20 to 30 years, maybe, for the technology to be available. Another 10-15 for the red tape.
Someone born today could still have a good career as an airline pilot.
People confuse machine learning and AI all the time. You can have bots that have seemingly meaningful conversations, and identify images easily. But they dont have any of I in the AI.
20 to 30 years, maybe, for the technology to be available. Another 10-15 for the red tape.
Someone born today could still have a good career as an airline pilot.
People confuse machine learning and AI all the time. You can have bots that have seemingly meaningful conversations, and identify images easily. But they dont have any of I in the AI.
Real AI is non-deterministic by definition.
Deterministic algorithms (scripts essentially) are certifiable.
Non-deterministic processes are inherently non-certifiable... only way it ever gets approved is by deploying it on a large scale with human crews as a babysitter and gathering literally years of operational data. You can never, ever be assured it won't screw up big-time tomorrow, but with years and many millions of operational flight hours you can reasonably assign a likely worst-case failure rate. So no exact science, at some point just going to have say "it looks pretty good so far" and hang it out there. Bureaucrats will be in no real hurry to do that.
The certification challenge is one of several very significant hurdles... IMO anyone old enough to read this post will have the opportunity for a full career, at pax airlines anyway.
#85
In a land of unicorns
Joined APC: Apr 2014
Position: Whale FO
Posts: 6,470
Yes.
Real AI is non-deterministic by definition.
Deterministic algorithms (scripts essentially) are certifiable.
Non-deterministic processes are inherently non-certifiable... only way it ever gets approved is by deploying it on a large scale with human crews as a babysitter and gathering literally years of operational data. You can never, ever be assured it won't screw up big-time tomorrow, but with years and many millions of operational flight hours you can reasonably assign a likely worst-case failure rate. So no exact science, at some point just going to have say "it looks pretty good so far" and hang it out there. Bureaucrats will be in no real hurry to do that.
The certification challenge is one of several very significant hurdles... IMO anyone old enough to read this post will have the opportunity for a full career, at pax airlines anyway.
Real AI is non-deterministic by definition.
Deterministic algorithms (scripts essentially) are certifiable.
Non-deterministic processes are inherently non-certifiable... only way it ever gets approved is by deploying it on a large scale with human crews as a babysitter and gathering literally years of operational data. You can never, ever be assured it won't screw up big-time tomorrow, but with years and many millions of operational flight hours you can reasonably assign a likely worst-case failure rate. So no exact science, at some point just going to have say "it looks pretty good so far" and hang it out there. Bureaucrats will be in no real hurry to do that.
The certification challenge is one of several very significant hurdles... IMO anyone old enough to read this post will have the opportunity for a full career, at pax airlines anyway.
#86
OEM's may need to incentivize the cargo operators to beta test the technology... from what I know of, it's not going live with pax until there's a whole lot of real-world experience.
#87
Gets Weekends Off
Joined APC: Mar 2015
Posts: 1,107
#88
In a land of unicorns
Joined APC: Apr 2014
Position: Whale FO
Posts: 6,470
#89
Gets Weekends Off
Joined APC: Mar 2015
Posts: 1,107
But how does it go? “Boxes don’t complain”
#90
If your plane flies a lot, better to pay a high acquisition/lease cost to get the best efficiency possible, which will pay dividends anytime the plane is moving.
If it sits a lot, better to go with older, cheaper, or even second-hand equipment. The less it flies, the lower the ROI on the tech.
That would apply to pilot autonomy too in theory, but in reality that will skip the pax airlines until it's been proven elsewhere... and cargo is the obvious "elsewhere" since mil technology will always be a hybrid man/machine approach since human creativity and flexibility is indispensable in warfare.
Thread
Thread Starter
Forum
Replies
Last Post