Self Flying Airbus
 

https://us.yahoo.com/news/airbus-sel...120900008.htmlAirbus' self-flying plane just completed successful taxi, take-off, and landing tests, opening the door for fully autonomous flight...Also: https://www.businessinsider.com/airb...-boeing-2020-4

It'd be interesting to see how many passengers would be willing to fly on an aircraft with no pilot onboard.

Most of them if you give them cheap tickets..........

I guess that would depend on the price.

Now find an insurer to insure not the plane but the passengers and third party liability, don't hold your breath. The control of the aircraft will have to be impervious to interference by terrorists as well.

What happens after an accident and you can’t blame the pilots?

The article doesn't appear to suggest this was a self-flying exercise. Instead it appears to be a automated refuelling exercise and the article mentions the autonomous flight earlier in the year in the A350XWB.

Like a dog walking on it's hind legs, nobody doubted it could be done...

Throw in a non QRH problem and it will probably not understand and crash

Pilot error is already the most common cause of air transport accidents and is trending upwards as mechanical reliability improves, technology mitigates ATC errors (TCAS/GPWS/GPS etc), and safety management systems bear down upon maintenance errors and organisational factors. And many posters on here routinely decry the standards of training, the experience, the pay and the working conditions of the younger and/or foreign members of the profession.

Far from being reticent over full automation, I’d be surprised if insurers aren’t looking forward to it, if not actively investing in research to help bring it about. The Miracle on the Hudson, the Gimli Glider, and no doubt a couple of other notable human ‘saves’ (*) make us feel good about what we can do that machines can’t, but ultimately they’re consolation scores in what’s going to become an increasingly one-sided contest as sensing and computing advance.

* Sioux City... but artificial intelligence nowadays can ‘do and learn‘ quickly enough that I suspect it could teach itself to fly on differential throttle just like Al Haynes did.

I wonder what it will say deep down in the fine print when flying as pax on one of these craft. Probably similar to your typical Windows operating system EULA where you absolve all your rights and accept the uncertified workmanship given to you as the norm forever.

Now if only all those software engineers livelihoods were on the line with jail time for negligence, only then will you see a real improvement in quality before release rather than a "it's fixed in version 2.0" response.

Tesla gets around this by requiring the driver keep their hands near the steering wheel - which isn't autonomous, though I accept designing for road travel has considerably more challenges.

I am a loss why we still keep throwing millions if not billions of dollars at a task that is relatively simple enough when the actual benefit in real terms proves to be quite marginal.

Of course it can be done, and unlike the dog walking on its hind legs, it can be done well. Given the level of automation available in modern aircraft, it's much easier to do a self-flying airliner than a self-driving car, because the environment is so much more controlled.

To me the puzzle is why Airbus are spending so much money on this.

Scenario 1: fully autonomous — nobody in the cockpit (maybe no cockpit). Remember the old joke about the automatic announcement saying "Nothing can go wrong ... go wrong ... go wrong ..."? However cheap the tickets were, I don't think it'd survive the first crash – and there would be a crash, because the world is imperfect and AI isn't really that clever. Murphy's law says something unexpected would be certain to happen that the automation couldn't cope with. Then parabellum's point about liability kicks in.
Scenario 2: supervised autonomous — pilot lines up for takeoff, presses a button and goes to sleep until the aircraft either lands or wakes him/her up because something odd is happening. Just feasible, but puts back most of the cost because you've got to have a fully-functioning cockpit with at least one competent pilot in it.

I agree with your statement that pilot error is now the leading cause of accidents in aviation, however what this statistic doesn’t account for is how many times pilots intervene in ‘minor’ incidents to stop the aircraft putting itself into a dangerous position.

Personally, I’ve seen an Airbus as it goes from ALT* to ALT take managed speed to MACH 0 and take all the thrust off, no explanation for it. I’ve had a DUAL ADR FAULT where the aircraft becomes fairly useless in terms of protecting itself. These are small examples and I’m sure there are many, many more. I also accept that in time these will be ironed out, but as long as Airbus are releasing OEBs and Boeing are installing faulty Alpha Protection systems, that pilots are needed as a backstop. We are faulty and full of latent errors, but at the moment everyday we make minor corrections to the automation that keep the aircraft safe.

Do I think aviation will be automated eventually? Yes. In the near future? No.

You'd certainly want to know how it would troubleshoot a non-standard problem - presumably there will be a data-link backup to allow input from a ground station in such cases (but this could not be guaranteed).

That said (and playing devil's advocate to some extent) it would be considerably less likely to make a total hash of a straightforward failure (or no failure at all) resulting in the destruction of the aircraft as has happened with human crews on more than one occasion! It's also not going to spill coffee on the avionics panel, drop its camera onto the sidestick or let its nephew have a go at the controls. (It won't try to pull the cabin crew either - although since it's French that will probably be an optional add-on! ;))

That last sentence at least gives us something to smile about on a bleak subject!

Cheap tickets shouldn't be problem then, right? You save the costs for the cockpit crew...

I am sure they will find someone. Engineer? Maybe roboots and electronics have emotions and are humanized through artificial inteligence until then. So robots on court? hahah

immediately call bullsh1t on that one but even if it were true, Hal wouldn’t have the local knowledge to put it down at an airfield that wasn’t in the database..

I think you’re mixing Sioux City and Gimli there; like I said, Gimli is definitely one up for the human pilot. Helluva deficit to overcome though. On the wider principle, local knowledge isn’t part of the licensing standard for any air transport operation (except for airport-specific qualifications). Can you imagine how hard it would be to learn and retain if it was?! London black cab drivers have to learn every street and every short cut in a few square miles... and their accumulated professional knowledge has been rendered pretty much irrelevant by GPS, traffic monitoring systems and Google Maps.

The myth of Pilot Error
 

Easy Street, WonderBus

Myth - an assumption about something as taken for granted rather than verified.

Cause - a verified hypothesis.

Pilot error assumes that you can both form and verify a hypothesis for human behaviour in all situations; which to date we are unable to do.
Thus humans can never be a 'cause', only a contribution to good or not so good outcomes.

The issue of human behaviour would also apply to design, build, and maintenance of the 'automation', thus autonomous operations with a higher level of safety than current operations may be an (unachievable) ideal.
Alternatively combine man and machine as a system which makes use of the best of each in normal and non normal operation.

https://static1.squarespace.com/stat...ech-Report.pdf

http://www.iploca.com/platform/conte...afetyMyths.pdf

Presumably, once the vending machines are installed, it will be totally autonomous. Sorts out the hijacking problem as well.

MD 11s were experimentally flown with PCA, Propulsion Controlled Airplane and it worked well but there's already hydraulic check valves that almost completely militate against another total hydraulic loss

The cockpit crew accounts for about 1.5-4€ per hour of flight per passenger. The numbers are approximate but it gives an order of magnitude.
So your low cost flight could go from 45€ to 43€, your long haul flight from 1000€ to 970€.

So yes lower price could mean more success but I'm not sure it's guaranteed, I hope it would not be !
Because I have at least a little hope that people will not favour products that destroy jobs.

Also because launching a new generation aircraft is always more dangerous at the beginning.
Interesting, but what about decision making ?
You see this side of the coin because the failures of automation that are corrected by human pilots are not tracked and their numbers publicly revealed.
But as a former accident/incident investigator, I can tell you that technical failures that are very easily solved by pilots are about as common as flying errors.

Blame the programmers, obviously!

Joking asside, autonomous flying is easier as autonomous driving. And AI can flawlessly execute checklists at least. And isn't subject to drinking and somatogravic illusion. I doubt it can resolve a previously unencountered parameters though, such as uncomamnded MCAS input. But between the children of the magenta and AI... in the next 10 years, I think I'd go with AI as pax. Also, don't forget that cargo doesn't care about pilot type.

You don’t need AI to execute a checklist; any computer can do that flawlessly right now. AI is what you need to go beyond the checklist.

No pilot = no minimum stick force gradient needed for certification = no MCAS... but anyway, AI would spot the trim position running away and take action. Here’s why I think statements of flaws in current systems (eg the Airbus speed setting itself to zero on mode change) are missing the point a bit. AI isn’t likely to be implemented by modifying existing aircraft systems. Rather, it’ll probably be a separate ‘entity’ with monitoring and input functions, interfacing with other systems, interpreting what it sees and taking action accordingly.

I've seen this too, but not for a long time. I think it happens when you put in a time constraint, then go direct to another point beyond the constraint or something like that. It's been ages to be honest.

Whilst a pretty serious bug, it's perfect for machine learning to solve. 'AI' (big data) will see that a pilot has had to make an intervention (selected speed) and then in future, know that that is the correct course of action. What manufacturers should be doing (and as I understand it, at least Airbus are exploring) is filling the aircraft with incredbly fine detailed monitoring to generate data on how pilots deal with scenarios; weather, abnormal and routine decision making. The data will be there, but it is just too complex to interpret at the moment.

You're expecting a lot, a sudden attitude change. Asking people today "would you go on a plane without a pilot?", obviously ALL will say no. The same would have been true in 1903 if you asked people would they trust Orville to take them for a ride. It will be a development towards that over a long time period. It would be naieve to say that machine learning could never do a better job than humans.

Do ATC and Tower become redundant? What if a tug wanders onto a runway while a plane is landing? What about stack-ups in weather delays?

Good post and agree.

Its quite easy to imagine problems computers dont handle well by themself if no one did the software telling them what to do. I think the US Airways 1549 /Hudson wouldnt have been that miracle without pilots.

Probably pilots save more situations than crashing aircrafts.

Back in the seventies, I contacted the local Professional Engineer licensing body about registering as a Computer Engineer.. They replied that there is no such thing.
So I went on to practise as such for forty-five years, eventually passing on my knowledge and techniques as a college professor to younger folk.

A couple of years ago, I met the President of the PE Association at an alumni event. He laughed and said they were still working on that accreditation.

People who design bridges and buildings have to be licensed, but not people who design computer systems that can kill hundreds when they go wrong!

BTW, if you have doubt as to my abilities, I tell people, you should not live within 26 miles of any Nuclear Generating Station where I have worked on computers!

They are, on their test flights. Each and every lights records each and every parameter and all that is stored in a data lake.

No, it's not.

Also, don't forget about the military.

Talking about costs for average short haul - each passenger pays in the order of £1 for the FO and £2 for the Captain.

£3 saving isn't going to amount to much - perhaps squeeze in 6 or 12 more seats depending on config if there is no FD but it's not an enormous saving.

I guess autonomous trains are currently the 'easiest' form of automated transport in terms of engineering and logistics.

There are currently only a handful of them operative and on relatively short tracks... I personally don't see aviation going automatic before the railroad system, but it is my mere crystal ball.

Pilot/operator error is a symptom, not a cause. The tasks usually automated first are those which humans aren’t good at,boring,tedious,often requiring high levels of precision and repetition. Dull and precise isn’t our thing, operating at peak efficiency, we are still prone to error, now add fatigue,stress,overload,illness and our error rate increases.
Machines are getting better at the abnormal, novel and unusual and will continue to improve.

We continue to repeat accidents in the same way.Collectively as Mark 1 humans, our performance is not and will not, noticeably improve.
Ultimately there will be fully automated pax aircraft, this is pretty much a given, whether a single pilot/machine minder remains In the short to medium turn,will be driven by safety, the market and PR.
There will of course always remain a need for engineers. As the saying goes,’you can teach a monkey to ride a bike, but you can’t teach the monkey to design or fix it.’
Unfortunately the monkey is now ‘optional’.

Already happening. Garmin has released 'Safe Return' (Cirrus) or 'Autoland - Halo' (Piper) which , on button press by passenger, takes over the rest of the flight including weather, navigation, and ATC, lands at nearest suitable airport, and opens the airplane .

However I wonder how it would handle the Miracle on the Hudson scenario.

Something a lot of people have overlooked is the performance aspect. Granted, it won't work for anything other than freighters, but no pilots = no real need for pressurisation unless you are carrying livestock. No bleed air demand = a what, instant 10% improvement in engine performance. Longer engine life, higher ceilings, or lower fuel flows for the same flight envelope. No pressurisation requirement = a lighter structure for a given strength, and/or no airframe cycles with their associated maintenance requirement.

It'll be a loooong time before we see it with passenger aircraft, but I would be genuinely surprised if I don't see autonomous freighters in my lifetime.

HAL would initiate immediate turn to the airport and land on a runway.

Perfect scenario for the airline. They get to blame the manufacturers

Once the 40 and up age group dies off there won’t be much reluctance to fly on automated aircraft by the public. The younger generation is much more trusting of automation

I think that one of the major issues in the short to medium term with ML/AI in aircraft will be that of certification. The regulations surrounding anything electronic that could have an effect on the flightpath are lengthy and complex. It’s one thing approving a documented, single-threaded application written in a “safe” language running on well-known hardware (and that can take years). Replace that with a black box that seems to do the job most of the time but no-one really knows how and you’re going to have to re-write the rules comprehensively.

This is not to say it won’t happen eventually but there are parallels with self-driving (level 5) cars and fusion power - it’s always some time in the future and when you get there, it’s more difficult than expected and the date gets pushed into the future again.

That's the impossible turn... doesn't work in any airplane...you can not make it, nothing or no one can. You will crash!

Hope the airlines keep some bitcoin laying around or the inevitable hack job.

https://www.dailymail.co.uk/news/art...e-systems.html