Aviation Week & Space Technology

Podcast: Pilotless Airliner—How Realistic? How Soon?

Aug 11, 2017

Pilotless aircraft have been around for years. The future will no doubt include large airplanes—major cargo carriers and eventually passenger aircraft. How soon will it happen? How will it work? Executive Editor Jim Asker leads a discussion with Graham Warwick, managing editor for technology; Michael Bruno, senior business editor, and John Croft, avionics and safety editor.

Don't miss a single episode. Subscribe to Aviation Week's Check 6 podcast in iTunes.

   

Discuss this Video 29

on Aug 14, 2017

Where do I donate money for you guys to get a decent mic?

on Aug 14, 2017

That will happen about the same time trains run without engineers!

on Aug 14, 2017

80% of crash are pilots related...ok but as an airline pilot, I never saw statistic showing how many crashes are avoided by pilots disconnecting, reseting, or realizing database problems or such computer issues.Sometimes having to take-over manually...as nothing works anymore.
It's the daily business today and I will take the train, even if automated instead of a fully automatic airplane for a while !

on Aug 14, 2017

It will eliminate pilot error as the cause of the accident in the NTSB reports.

on Aug 14, 2017

They will just change it to "Programmer Error".

on Aug 14, 2017

I guess that means starting today!
Last summer I rode on a Paris Metro with no driver at all. I sat in the front looking out the front window. And it wasn't a new line but one of the original lines (I think line 1) that was converted to driverless operation.
Also, driverless freight trains have been running for some time in Australia and on some short lines in Arizona.
I'm guessing within the next 20 years train engineers will be completely phased out and pilots shortly thereafter.

on Aug 14, 2017

Given that switching is done from a control center, trains move along a one dimensional path, almost impervious to weather. A computer pilot will be millions of times more complex. Stopping and waiting for updates or repairs will not be a choice.

on Aug 14, 2017

I took my kids to Disney land about 30 years ago and we sat up front in the Monorail without a human driving it.... many of the 'trains' people take to the gate are already autonomous. Trains are dead easy by comparison (stop/go) and can be programmed conventionally (even with simple electrical relay systems) whereas airplanes and cars will take AI/Machine Learning systems but that's just a matter of time.

on Aug 19, 2017

There are many trains without engineers. There have been for decades.

on Aug 14, 2017

Really interesting. My home airport (as a passenger) is subject to notorious windshear but I recall nearly 15 years ago a pilot telling us he was landing on Autopilot in a 767. Totally successful landing. I also recall my fascination at autopilot guidance 30 years ago. I also recall landing in horrendous conditions where I am sure the human intervention got the plane down on the runway safely. Not sure how confident I would be without a potential pilot override - but sure it will come

on Aug 14, 2017

Airbus, who are probably leading in FBW highly integrated flight controls (sorry Boeing) could not even program an ECAM to manage all the potential indications after a Rotorburst. Flight crew abaility alone save the AC. if they could not manage even to present coherent and non distracting data for this one, reasonably probably event, what hope do we have that they could identify all potential hazards and combinations of hazards (including dormant faults/acceptable deferred defects) and write computer code to manage, control and recover the AC. Its close to impossible with the number of combinations which would be possible. Same for AF 337. So many FCS systems depend on valid airspeed, how does the SW cope when all are misleading. Getting fully automatic cars which can pull over and stop safely will be enough of a challenge, Getting an AC to recover after a major (not extremely or even remotely improbable) failure or combinations of failures, well thats a job for life. Even getting UAVs to fly in shared airspace (see and avoid) was uphill.. Im glad I will have passed on before this comes to pass.

on Aug 14, 2017

David, the Airbus systems are traditionally programmed (I.e. the programmers needed to think of all of the possible scenarios and program in the right behaviours). This is NOT how the newer Artificial Intelligence / Machine Learning systems work at all. Those systems are "trained" much like humans are and use a combination of pattern recognition and statistics to determine the right course of action based on prior history of similar scenarios. It's a complete paradigm shift in computer programming and not many people (even programmers) actually understand it that well yet but perhaps a very simple example can help shine some light. You can show these systems 50 or 60 pictures of a type of dog, you don't need to tell it that it's a picture of a dog or anything but let's say you say it's a "Collie". You could show it then another 100 pictures of baby's, cats, cows, cars, and generally the system will do as good a job of classifying Collies as any human. And as it sorts through those pictures, it takes note if it was right or wrong and "adjusts" to get it better the next time. After about 1000 pictures, it will generally never make a "mistake" again. It takes about 5 lines of Machine Learning "code" to do this - not much more complex than here are a bunch of pictures of Collies... now go find me the Collies in these pictures, if you guessed wrong, I'll tell you. To try to do this in the traditional programming environment would take tens of thousands of lines of code and would likely not work very well anyway - and as you pointed out, if the picture was blurry or something, it might make a mistake because the programmer didn't account for that.

Same thing for driving a car down a road... the way they do it is by generating lots of images in real time (plus with hundreds or thousands of cars networked showing photos of the roads over months/years), they record the action the human took (how much they turned the wheel, what speed they went around the corner, putting on the brakes to slow down, etc.) and the computer systems watch and compare the response they would have taken to the action the human took. Over time they become as good or better at it than humans (they don't drift off or get bored and use their iPhone's).

These systems will be in place "watching" real pilots for years but as the autopilots get widely deployed, because they are networked (and/or can upload their learnings after landing), they learn concurrently and can very quickly become expert. UAV's don't (yet) do much of this... you need yvery fast processors (typically many graphical processing units all working in parallel) and drone's so far at least just don't have the "umph". Of course that's a matter of time. The amazing thing about ML systems are how fast they "learn" and how complex a task they can handle without human intervention. Google GM autonomous car in San Francisco to see what was possible 2 years ago. It's not if, but when...

on Aug 14, 2017

Nothing can go wrong, go wrong, go wrong.........
Sullenberger and Asseline (1988) can intelligently testify about computers being in charge of an airliner rather than a Captain.

on Aug 14, 2017

Not putting down Sully but a computer could have figured out it didn't have the altitude and/or airspeed to make Teterboro in a fraction of a second... it took Sully hundreds of thousands of times longer - the FAA pilots who tried to recreate what Sully did failed many times so it is not a given that humans could always have saved the day like Sully did.

I'm sure that there were lots of elevator operators saying the same thing back in the 20's. If cars can drive the rush hour traffic of downtown San Francisco without human intervention, they will be able to fly airplanes someday. It's inevitable.

on Aug 14, 2017

The catastrophic damage to the aircraft after an engine blew it self apart and caused major damage to the wing and the hydraulic systems was only saved by the crew who were able to think outside the box. No algorithm could have foreseen all the problems the pilots had to deal with and eventually saved the aircraft.

on Aug 14, 2017

“Probably”? “May”? “But”? “Almost Certainly”?
Your opinion of pilots is as high as that of analysts???
You guys need to give your heads a shake.
Technology has complicated our profession as much as it has added value so be very careful in your assessments of the need for skilled Airline Pilots.

on Aug 14, 2017

The idea that a "nuts and bolts" code writer could match the knowledge of an experienced pilot is --- ? I've dealt with this third-party software stuff forever. The programmers can't know more than they're told by the customer. After the programmers are instructed by management, who know nothing of the specific tasks, they write inadequate instructions. Then comes debug. If you haven't thought of a possible situation, you can't program how to handle it. If you don't know exactly what is needed, you can't program for it. We've got a long way to go.
On the road, I know that many of my fellow drivers are inept. I'm ready for almost anything. Almost. How much code would that take?

on Aug 14, 2017

That's not how these systems are "programmed"... they learn like humans... watch repeatedly, come up with a prediction, compare it to the correct action, repeat thousands of times get better and better over time and then actually start to take control. They are NOTHING like the old programmatic computer systems of a few years ago. You'd be surprised at how little code it actually takes... 50 lines can keep a car between the lines and from hitting obstacles (cars, bikes, etc) ahead.

on Aug 14, 2017

Not this guy. Do you really trust people to produce this, who charge $1000.00 for a cell phone made in China? Don't worry the government will make sure it's safe-HaHa! They magically tuned crop-dusters into bombers!

on Aug 14, 2017

I was fortunate enough to be part of the development and testing of an 'Optionally Piloted Aircraft' a decade ago. There is no question that a dialed-in control system can outperform a sack of bloated protoplasm any day of the week. However, the fun begins when you lose link, have a critical sensor failure, or depend on vulnerable satellites or condition-specific onboard sensors in conditions that are 'out of bounds' for landings. Programming for improvisation may be possible with AI, but until we can replicate an instinct for self-preservation and the ability to see things from a human perspective, fllght crew replacement as a goal is misguided.

on Aug 14, 2017

Transition, transition, transition. Both common sense and the public's anxiety will be served by an unrushed transition, backup pilots until two things happen. A failure rate of the A.I.s approaching zero, and the acceptance by the public. Could take a while. Don't rush it!

on Aug 14, 2017

I am not sure I would get into a driverless car.... let alone a pilotless airplane....

on Aug 14, 2017

It already feels, on certain airlines (which I would not name due to legal reasons), that the cabin crew members are already robots. Having the pilot be a computer would not make such a huge step.

on Aug 14, 2017

So, I'm picturing a test airliner with the cockpit equipment relocated to the cargo hold and picture windows in the nose... and lots of members of congress on board.

Oh that wasn't very nice. Disregard.

on Aug 14, 2017

As an "older Aviation Electrician" I can understand where the industry is going. However, Having an experienced Human in control, often means arriving alive! Capt Sully, and Landing a passenger jet in the river. Or bringing home a F-4 to the carrier; after a mid air with a Soviet aircraft inside the Artic Circle. Never underestimate a motivated, invested human.

on Aug 14, 2017

Started in General Aviation..then 37 years with the Airlines..

The Podcast actually was very informative. Yes, as I said before, this will happen.. I just think its further away then the 'bean counters' are projecting.

All or most of us have a very different view of this as we have spent many hours flying all kinds of aircraft.

Yes, Autoland has been around for sometime now, but it still has to be programed by a Pilot. They seem to work much better when new.

Additionally, I was a Navy Electronics Technician on the Grumman A6A.
I know, I'm old, but I have very good prospective on the advances in automation.

In my opinion, General Aviation will most likely be last to lose pilots as the job is more diverse. In many ways their job is a lot more challenging than an Airline Pilot. Possibly FAR 135 freight first, then FAR 121 freight, but anything passenger will be last..

I do like the idea of placing 'programmers' on these flight for the first 6 months , then random flights with their families.

Robot FAA Inspectors ? Guess they need to go back to school for some programming lessons.. I'd rather drive my RV !!

on Aug 19, 2017

I agree. The pilotless airplane is a simpler problem than a driverless car. Autonomous airplanes are already possible. Once people become accepting of driverless cars the job for airline pilots will rapidly dwindle.

on Aug 15, 2017

Glad that the airline business is in search of ways to solve the problem of fatigued pilots. You will only need to change batteries...and of you go to the big blue yonder...
In the meantime , in a galaxy far far away........

on Aug 17, 2017

I reject 100% of the purely intellectual endorsement(s) of pressurized aluminum tubes, full of passengers, flying at 30+ thousand feet over oceans.
There is no narrative which eliminates the myriad concerns regarding unplanned contingencies, emergencies, anomalies, passenger issues and hundreds of other 'unknowns'.
I'm willing to evaluate and consider a single pilot scenario - but as long as I'm above ground - there will be no 'pilotless' flight for me.

Please log in or register to post comments.