Airbus' self-flying plane just completed taxi, take-off, and landing tests | Hacker News Reader

Airbus' self-flying plane just completed taxi, take-off, and landing tests

businessinsider.com

332 pointsapsec112318 comments

A friend of mine, a military helicopter pilot told me years ago, that Airbus can basicaly fly by itself, except for taxiing. It was about 15 years ago.

But then the airline prohibited to use the automated systems during take-off and landing, because the pilots don't get practise. But pilots need those practise if they should take over in emergency situations.

The same is true for automated driving. I always say, we will see automated railways and automated airplanes before we see automated cars.

PS: Because of some comments, Nuremberg had the first railways system to allow mixed traffic of automated and manual railways. Yeah, I know some of the guys doing the saftey certifcations for this.

> we will see automated railways

The future is now: there are automated rail lines in a numbers of cities around the world. Some are relatively old, too.

The JFK AirTrain is one such example.

Most of the automated railways operate on closed tracks. Many of those even on closed platforms. Here in Germany, my hometown Nuremberg, it runs one of the first underground railways with open platforms and combined with man operated railways. It needs a lot of safety electronics and I know some guys who were responsible for certification after install. (Yeah, I am somehow from the business.)

Don't try to compare closed tracks to open tracks. A closed track is more like an elevator.

We have the London DLR

The Docklands Light Railway is both grade-separated (which I guess is what your parent was getting at with "closed"? hard to tell) and only GoA (Grade of Automation) 3.

There are GoA 4 systems in use in the world. The difference is that every DLR train has a trained operator onboard whereas in a GoA 4 system the nearest employees are in some distant operations centre, watching only via a CCTV connection. GoA 4 trains are entirely autonomous, if one were one day to somehow break down so badly as to be inoperable a separate team would need to go rescue it and meanwhile other trains would simply stop at a station and allow their passengers to disembark.

He compare "closed" to elevators, so I thought that "open" meant no junctions, but you can be also right.

Usually there is no operator onboard. But is still GoA 3 because emergency situations are handled by operators from a control room.

There is always a trained operator on the DLR.

Look around, you will see there is a person wearing a uniform who is perhaps helping tourists figure out where they need to change trains, or assisting a woman struggling with shopping and a push chair. These are TfL employees trained to drive the train, even though it's autonomous.

Before the train leaves, even under normal autonomous operation, this person in the uniform activates a "close doors" control and this is the only reason a DLR train will close its doors. When the doors are closed and the way ahead is clear the train will drive itself to the next station, and open the doors, the cycle repeats.

The person wearing the uniform is trained to safely operate a train. Any time they're on first shift they actually sit at the front (usually occupied by excited children) and unlock the onboard controls to literally drive the train for route verification purposes. Autonomous trains aren't looking, and tools or equipment may be left on the track by work gangs (despite a "count it out, count it in" procedure) so the first trip of the day is always done manually.

You can't drive a DLR train from the control room.

What are the main difficulties around an open system vs a closed one? To my novice eye they look like they should be fairly comparable.

Runs on my laptop vs runs in production

A semi truck breaks down in a position fouling the tracks. A drunk passes out on the tracks. A freight car isn't securely tied down on a siding and rolls out onto the tracks. Et cetera.

For an open system, you have to be able to respond to arbitrary visual input. For a closed system, there shouldn't be random stuff on the right of way.

And that we don't have AI that can handle these relatively easy scenarios is soft proof that we are nowhere near the hard scenarios of fully antonymous driving.

For most train situations, isn’t it impossible to stop once you’re in visual range anyway?

Depends on where you are. Some places are flat enough that you can see several miles. A freight train can stop in two miles, more or less. You're not going to see the drunk on the track in that range, but you might see the semi or the freight car.

Think outside the box, the camera could be mounted at the intersection instead of on the train. I'd presume whoever is certifying this stuff also ensures if the gate isn't verified to block traffic, the train stops before the intersection

In Britain these are named MCB-CCTV ("Manually controlled barriers, CCTV monitoring").

At the crossing are boards indicating that this is a railway crossing, two sets of lights for each direction of traffic (two amber, one red in each set) and barriers, typically long motorised boom arms from which a flimsier but more visible structure hangs, plus a tall post with CCTV cameras mounted on it. The CCTV feed goes with all the other signal data to either a "signal box" not so far away, or a larger centre where many signallers work together. A signaller who wishes to move a train through the crossing views the CCTV feed, and initiates a sequence in which the lights show alternating amber ("caution, stop if possible"), then red ("stop"), the barriers close road lanes entering the crossing, then barriers close the road lanes which exit the crossing. The signaller verifies everything is as intended (e.g. no vehicle stranded on the crossing, no idiot pedestrians clambering over the barrier and standing on the railway) and then authorises trains to pass. Once as many trains as necessary have passed, the signaller closes the crossing to trains, and tells the crossing to re-open to road traffic, it raises the barriers and then switches off the red "stop" lights and road users can now use it as normal.

Crashing at 20mph is better than at 80mph

Passenger trains are far lighter than cargo trains.

I'd take Amtrak to my hometown, and the engine would blow past the platform at 20+mph and still stop the passenger cars right on the platform. Riding on the train and it seemed like they'd miss it.

Afaik the breaking capability of cargo trains is lower because they use simpler brakes and only use the airline for breaking, which limits the speed at which you can apply the brakes on a long train without risking breakage of the links. Passenger trains use disk brakes and through electronic control can apply the brakes evenly across the entire length of the train, so the brakes can be applied much more quickly.

I think the weight of a car itself wouldn't change the theoretically possible stopping distance, since the momentum is proportional to the mass, but so is the friction between wheel and track, so mass cancels out.

The drunk is dead. There's no great risk to the safety of the train from striking a human or small animal (larger creatures like bears and deer can be a problem). There are a number of very dangerous places to pass out drunk, and it feels like that's not really a problem for railways to solve.

At-grade ("Level") crossings should be closed (and thus grade separated). Where that's impractical there are numerous options for ensuring the crossing is not open to railway traffic until it is clear:

In some areas it can be practical to employ a person whose duties include verifying that this crossing is clear and then authorising its use. Because CCTV is cheap now we can fit cameras and lighting and have them do this remotely for a large portion of the network.

Or we can use radar to protect crossings. The barriers drop, and then radar is activated (the small hazard introduced by using radar is mitigated because the crossing is closed, nobody should be in the area being illuminated by the radar anyway), if somehow there is a large object on the crossing even though the barriers are down, this means something went wrong. The outward flow barriers re-open to potentially allow a trapped driver or animal to leave and the system tries again, if this doesn't help the crossing is considered failed and humans go to investigate, they will find your broken down semi-truck.

The freight car should preferably be harmlessly derailed by trap points. If this is impossible (and the layout cannot be rectified) the presence of the freight car should trip sensors, either axle counters or track circuits, and close the section to traffic. Again humans will likely have to come investigate how that occurred when it cannot be explained and won't go away.

I would be very surprised if hitting even a bear could cause a problem for a train. Light rail, maybe. Heavy rail, or a freight train? I doubt it.

Perhaps “open” and “closed” are synonymous to “conventional” and “purpose-built” in the industry?

Purpose-built automated lines often use smaller cars with more powerful motors sometimes concrete tracks. I suppose those help computers in flattening acceleration/deceleration curves and eliminating needs to compensate for weather or passenger counts.

What are the drawbacks to automated rail? I had to look this up and was very surprised rail is not automated more in America. Is it safety?

In BART, in the bay area, I believe it is largely because of Unions.

Well, BART had one crash because the automated control system mis-read a command and accelerated when it should have been braking. Debuggable? Perhaps. But after that, it's a tough sell to say that there shouldn't be a human in the cab to say "Wait a minute, that's not right."

Unfortunately there has been a prolonged campaign(ennui?) against rail in the US. I imagine lack of funding for innovation in legacy rail systems is to blame here.

Rail cannot get adequate funding for even basic maintenance and upkeep and upgrades, let alone modernization, because of widespread corruption among government officials and public sector workers. I'm sure public perception plays some role but it is not the root cause for our current situation.

A mix of politics, bureaucracy, and cost. For example, there was a fatal derailment back in 2017 in the US, and an automated safety system that may have prevented it was delayed.

https://www.bbc.com/news/world-us-canada-42421417

From the wikipedia article on Positive Train Control (PTC):

>In December 2010, the U.S. Government Accountability Office (GAO) reported that Amtrak and the major Class I railroads have taken steps to install PTC systems under the law, but commuter rail operators were not on track for the 2015 deadline.[14] As of June 2015, only seven commuter systems (29 percent of those represented by APTA) were expecting to make the deadline. Several factors have delayed implementation, including the need to obtain funding (which was not provided by Congress); the time it has taken to design, test, make interoperable, and manufacture the technology; and the need to obtain radio spectrum along the entire rail network, which involves FCC permission and in some cases negotiating with an existing owner for purchase or lease.[15]

https://en.wikipedia.org/wiki/Positive_train_control

So needless to say... the technology likely exists today, but implementing anything significant to existing infrastructure in the US is a monumental feat.

The above example is just a safety system, but once you start removing employees you also start running up against labor unions... which is yet another layer holding things back.

The legal liability issue. If you have a human driver, you can blame them add an individual for any incidents. If you have software in control, the company that runs the train or developed the software is liable.

American Freight automation will be a crazy beast. Closed lines can be fully automated, and we are seeing a lot of PTC (positive train control) such as automatically enforcing speed limits and collision detection. The driver is supposed to know when they are approaching a red signal -- nowadays the engine notices too.

However, the nationwide freight network is interconnected by many different companies, in various states of repair, with various states of mechanization. I live near a bunch of grain processors and when they are taking deliveries it is a lot of manual switching and back and forth to get each plant's cars to the right lines.

Could it be automated? Yes, the technology is available, you 'just' have to implement it. Will it be? Probably not anytime soon.

The metros in Lausanne, Switzerland and in Copenhagen, Denmark are automated.

Nitpicl: Only one of Lausanne's two metro lines is automated; the m1 is way too exposed to human interaction with the track area for that to be risked as it's currently laid out.

Also, Line 14 on the Paris metro (heavy use) has been automated since the 90s.

But we haven't seen automated freight trains. A light rail train hauling commuters along a dedicated bespoke elevated rail line (Vancouver's skytrain, Disney World etc) is one thing, a hundred cars full of iron ore descending a hill behind a preschool is another.

The preschool is hiding behind the hill on the railway line? I wish they'd stop building these schools on the train tracks, it's so irresponsible.

I believe that Australia has an automated freight line.

The original intent of the Black Mesa & Lake Powell (hauling coal to a power plant) was for it to be fully automated, but I don't know if it ever ran without a human in the cab.

Yep, the iron ore trains are largely automated these days, I believe.

>> has an automated freight line.

I didn't know about that, but is it a "freight line" or a "freight train"? If it is the former, a dedicated line used only by the automated train, then it isn't all that removed from disneyworld: a single-use train on a dedicated line. And Australia is notoriously flat.

I'm from the west coast, where driving a heavy train through the mountains is serious business, but disasters can happen anywhere. The latest epic disaster in Canada was caused by a train that was unmanned at the time. It was parked.

https://en.wikipedia.org/wiki/Lac-M%C3%A9gantic_rail_disaste...

Yes, I've been to Tehachapi, Cajon, and Donner Passes. And Cima Hill, and Ash Hill, and Goffs. I know what mountain railroading out west can be.

Maybe the gnarliest scenario I know of out there is: There's a major brush fire on Cajon. Like, 20-foot-high wall of flame. Is it far enough away for you to try to get the train past it? (With maybe 4000 gallons of diesel fuel under each locomotive, and maybe liquefied petroleum gas tank cars in the consist?) Or do you park the train, and hope the fire doesn't come to you? I mean, I guess an AI can make that decision without being influenced by concern for its own survival. But I suspect that most train-driving AIs don't have a really good set of training data for this scenario...

> I suspect that most train-driving AIs don't have a really good set of training data for this scenario...

I suspect most human drivers don't either. The AI might not make a particularly good decision, but is there any reason to believe that a human would make a better one?

There's a 1700km track network with both automated and non automated trains, so slightly larger than disneyworld.

https://www.railjournal.com/freight/rio-tinto-completes-auto...

"Australia is notoriously flat"

Citation needed.

REally? You need a citation to know that australia is flat?

"Australia is the lowest, FLATTEST, and oldest continental landmass on Earth[4] and it has had a relatively stable geological history. Geological forces such as tectonic uplift of mountain ranges and clashes between tectonic plates occurred mainly in Australia's early prehistory, when it was still a part of Gondwana. Its highest peak is Mount Kosciuszko at 2,228 metres (7,310 ft), which is relatively low in comparison to the highest mountains on other continents."

https://en.wikipedia.org/wiki/Geography_of_Australia

Russia is big. Switzerland is small. And Africa has gets less snow than Canada. (Citations needed)

The basic route of this train (Newman to Port Headland) is a gentle slope to the coast. +175 and -700m over a 400km journey. (Google maps) Compare that to a route in western canada: Canmore Albert to Salmon Arm BC, the same distance but +2500m and -3500m.

Being the flattest continent does not mean that the continent is flat. My point is that there is still plenty of hills and steep terrain around. It's not self evident that if you build a train line somewhere in Australia, that line will be flat. In fact the Pilbara automated rail network does contain some steep sections.

It still snows in Africa, and if you tried to climb Jebel Toubkal without a jacket I don't think knowing that Canada is snowier would provide you much comfort.

Vancouver has SkyTrain:

"SkyTrain is the oldest and longest, fully-automated, driverless, rapid-transit system in the world."

https://www.translink.ca/Schedules-and-Maps/SkyTrain.aspx

Which is a completely closed track, much like an elevator.

Not sure this meets your definition but the Expo line splits: https://en.wikipedia.org/wiki/Expo_Line_(SkyTrain)#Route

There's a few one that existed before then. Quite crazy to think that some existed in the 70's, almost half a century ago: https://en.m.wikipedia.org/wiki/Automated_guideway_transit

There are already automated trains in Paris (one metro line), and I can think of several places in Japan with automated train lines as well (not the main ones). It's already proven technology.

I would be very happy to see automated trains, I often see an immediate jump to tackling the hardest cases in engineering before thinking about simple and efficient solutions. Almost everything about the Japanese rail system is something I want to see in the US

Railway operation in Japan is indeed excellent and worth emulating.

However, busier services are very much dependent on human operators. Both on-board (2) and platform-side (half a dozen).

In London several of the tube lines are automated. There is a driver but all they do is press a button to close the doors at stations and make announcements to passengers "for safety purposes" (i.e. unions)

For some lines that go above ground (Northern, Jubilee, Central), the driver sometimes had to put it into manual mode to compensate for the worse breaking when it's raining outside.

This is a bit of a misrepresentation. Drivers still have actual safety and operational roles – they can drive the train manually in cases of equipment failure, deal with passenger emergencies, that kind of thing. Even the DLR, which has used automation since opening, has a staff member on each train who can take over driving if required.

That's not to say that entirely driverless services can't be done. There are a bunch of metro services globally that are grade-4 fully-autonomous and totally unattended, and some of them have been in operation for over 30 years. But like all large-scale systems, there are lots of complicating factors, system-specific conditions, and various tradeoffs to be made – a lazy Daily Mail "damn unions" argument isn't really fair.

Yep this is why they want "train captains" (kinda like the DLR has) rather than requiring a fully-trained driver in a cab at the front.

IIRC there is something like a 6 year waiting list to become a tube driver because they do a 36 hour week for £55k a year (up to £100+k for some) with over 40 days paid holiday, and can retire on a full pension at 50 years old (1)

It seems like a ludicrously cushy job and I 100% put it down to those "damn unions" essentially blackmailing TFL by always threatening to strike at the busiest times. I don't blame them,but I don't think that the "damn unions" argument is unjustified when it comes to Tube drivers and there near-constant threats to go on strike for the most rediculous things that any reasonable person would agree were legitimate reasons (such as when a driver was sacked for repeatedly failing drink-driving tests (2), or another for a driver who repeatedly drove through red signals (3)). These are potential disaster near-misses, yet the tube driver's unions use them as more leverage to threaten a strike and squeak out a bit more pay and a bit more holiday. And when they are not doing that, they're frankly taking the piss and threatening to strike over not being able to make a cup of tea (4)

1 - https://www.standard.co.uk/news/uk/tube-driver-salary-holida...

2 - https://www.telegraph.co.uk/news/uknews/road-and-rail-transp...

3 - https://metro.co.uk/2018/04/13/tube-workers-striking-protest...

4 - https://www.telegraph.co.uk/news/uknews/1337609/Unions-threa...

The wages and costs of train drivers are pretty insignificant in the costs of running the London Underground, so tiny a percentage of the running costs, yet people focus on them because they earn a relatively good wage (how dare they have a pension that might keep them fed in their retirement! Ban those greedy unions! etc etc ...) Shame on all you haters.

I just skimmed a few search results and the wage quoted here is within the upper end of a London average - depending on how you define it. There are obviously a lot of fudge factors and things that skew results due to various factors. Having seen and read of things that tube staff have to put up with, they seem to me to earn their wage. Below is a link to the story about the railway worker who died from Covid after being spat in by someone claiming to have Covid.

Note that staff appear to have been sent out unwillingly and were not allowed to wear masks.

https://www.theguardian.com/uk-news/2020/may/12/uk-rail-work...

>a 36 hour week for £55k (up to £100+k for some) a year

An ok-ish middle class salary. Not outrageous in its own right and i suspect if you compared it to the lifestyle of train drivers 50 years ago it might allow them to live the same sort of lifestyle?

It's no secret that our expectations and the job market have changed dramatically but it is telling that jobs with hard fought inflation adjusted wages have wait lists for those jobs. We're like a frog slowly being boiled and we need posts like the above that cause us to think "Hang on didn't train drivers of yesteryear support a wife and three kids? Could you do that on £55k?".

As tech workers we all likely earn so much that we could live the middle class lifestyle of yesteryear (paying off the mortgage before retirement, supporting a partner and kids, taking regular family vacations, etc.). I mean we like to think we're all super rich but really we're only where we should be. Perhaps we shouldn't be directing hate downwards here? The wealth gap above us has increased, whilst the wealth controlled by the bottom 50% is lower than ever.

The only way the disparity here can be corrected is to accept those below us in salary will fight for more. It's pretty understandable really.

It says a lot that you think all tech workers likely earn more than £55k.

Lots of bubbles to go around. Average entry level software developer job in Germany pays 40-45k€ (35-40k£). 60-70k would already be well-paid senior positions. No developer or dev manager earns six figures. These numbers also roughly apply to most engineering disciplines.

> An ok-ish middle class salary

Twice to four times the median salary in the UK (and three to six times the salary of a bus driver who is usually far more skilled), for a low-skilled job that requires neither education nor physical exertion. There is a reason why there are massive waiting lists.

I thought that on all Underground lines the driver has to operate a deadmans switch at all times - if released the train stops. Is that not true?

Used to be, and probably is still on some of the smaller lines like Bakerloo or Circle. But on the main lines like Victoria and Northern they're all automated now. Often you see the train coming in to the platform and the driver is just sitting there reading their phone/playing candy crush etc.

Same in Milan. The M5 line is completely autonomous (you can also sit in front and look at the tunnel through a large window). The next line, the M4 (no, not a typo) will be autonomous too.

I don't ever recall seeing a train driver in my trips on Singapore's MRT either now that I think of it. Is this relatively new technology? Some of the older trains on the North East and Downtown MRT lines are at least a decade old, if not older. It would seem like that such tech should be ubiquitous in most cities with a good public transit system.

> It's already proven technology.

Automated trains are just like elevators, except they move horizontally.

There used to be an operator on every elevator, but they have been replaced by electronics since a very long time ago.

But go further, faster and carry way more people while having an incredibly varied operating environment, unlike an elevator/lift.

The whole basis of signaling kind of makes it all moot. If the signal is red, the train stops. If the signal is green, you go through. Their may be yellow signals as well, generally meaning that there is train x signals ahead, proceed at caution (generally reduced speed).

A very large number of accidents are due to the operator failing to follow these very basic signals.

>incredibly varied operating environment

How many variables do you see here?

The absolute safest scenario is likely automated trains with a driver with a red button to stop the train, in case of emergency. After years go by without even having to use the button, even those operators could be removed. We're living in a world where automated driving is starting to overtake automated trains, something that literally exists on rail, and switching is controlled electronically on almost every passenger rail in the western world.

[0] https://www.washingtonpost.com/transportation/2020/07/07/red...

[1] https://en.wikipedia.org/wiki/Hordorf_train_collision

[2] https://en.wikipedia.org/wiki/Hordorf_train_collision

[3] https://en.wikipedia.org/wiki/Signal_passed_at_danger#Accide...

> How many variables do you see here?

Unless we are talking metro, where I live it is not that uncommon to have some animal wonder on the tracks, people trying to commit suicide on the track, people throwing things at the train from bridges, ...

Hell, I there was even an hot air balloon that crashed on the train track a few years back in my city, blocking trains for the whole day.

Those are really all one variable, object on tracks.

And I got some bad news if you think 140 car freight trains stop for a deer on the tracks.

>people throwing things at the train from bridges,

I'm not too sure what you think an operator does today, besides maybe calling their dispatch/911.

> automated trains with a driver with a red button to stop the train

This is an ergonomics nightmare scenario. The operator needs to pay attention to keep situational awareness all the time while just observing the train doing its thing. Then, all of a sudden, the train starts doing something that it shouldn't be doing (due to a failed signal, another train doing something stupid, an obstruction, a malfunction...) and they need to quickly react and do a safe abort (slamming the brakes may actually be a bad idea sometimes)

This is an issue in aircraft with heavy automation - everything seems normal until it isn't and then the pilots need to quickly figure out exactly what happened, how they got where they are and how to get out.

>This is an issue in aircraft with heavy automation - everything seems normal until it isn't and then the pilots need to quickly figure out exactly what happened, how they got where they are and how to get out.

This isn't flying though, and you can stop a train by slamming the breaks, and this is perfectly fine in 99% of scenarios (and which is exactly why each train car has an emergency break cord/button).

I can't really imagine a situation where a possible derailment is better than stopping the train. If your on a narrow bridge, surely stopping is better than the possibility of being unable to stop when the bridge is out.

>> Automated trains are just like elevators, except they move horizontally.

Really? I haven't heard of any elevators killing dozens of people lately. I haven't seen many elevators crash, burst into flames and effectively remove entire city blocks from the map. I haven't heard of people studying elevator operators for PTSD.

https://upload.wikimedia.org/wikipedia/commons/e/eb/Lac_mega...

"Traumatic exposure and posttraumatic symptoms for train drivers involved in railway incidents"

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4462444/

>>Because of involuntary exposure to PUT [Person Under Train] incidents, the likelihood of train drivers to witness the violent death of a person is much higher than that of the general population, and that puts the train driver at risk of psychological trauma.

>>Of the 193 train drivers, 152 (78.75%) reported at least one PUT incident. Respondents reported as much as 14 PUT incidents/person, with a mean of 4 (SD = 2.83) incidents/train driver.

In all honesty, a train driver is more likely to see someone die violently in front of them in a given year than the average soldier.

“Trains” and “automated trains” practically are two distinct concepts like “cars” and “golf cars”

An eggcorn maybe?

The vehicle you're probably thinking of is a golf cart.

A Golf Car is a popular model of motor vehicle from Vauxhall.

All the more reason to take train drivers out of the picture.

Doesn't solve the problem of people using trains to commit suicide, which isn't certainly going to be solved by any amount of automation.

So? Nobody said that it would, just that drivers won't suffer the associated ptsd.

Need some of the mitigations they use in Japan like calming blue lighting at the platform ends, full length barriers and gates.

> Automated trains are just like elevators

Ahem. No.

There is no collision hazard in elevators (essentially each elevator is on its own dedicated track). In railway, collision hazards are everywhere because you can change track and there are multiple trains on the same track, possibly (yikes) in head to head movements.

If you want to see the specs of a real Automatic Train Protection, the European one is public [1]. And that is only the protection part. There is no automatic pilot in there.

[1] https://www.era.europa.eu/content/set-specifications-3-etcs-...

Nitpick: there are collision hazards in some modern elevators. See for example https://www.popularmechanics.com/technology/a20847/the-futur... (company pages are, IMO less clear: https://twin.thyssenkrupp-elevator.com/ or even https://www.thyssenkrupp-elevator.com/en/products-and-servic...)

Vancouver BC has had a fully automated train system since 1986. When it started they were using VAX as part of the control system.

Toronto had the demonstrator line that saw and continues to see revenue service despite its ancient age. Toronto’s cars were built with a full cab, partly because of pressure from the union, to allow an operator to push the button to close the doors. (1)

The system no longer runs in ATO (automatic train operation) because of ongoing issues around maintaining contact with the mainframe when it snowed.

(1) https://transittoronto.ca/subway/5107.shtml

2 actually.

Line 14 was conceived as driver less. In 2012 line 1[1] was converted to fully automated. Others, to the best of my knowledge, are planned.

[1] https://fr.wikipedia.org/wiki/Ligne_1_du_m%C3%A9tro_de_Paris (In French)

There are ongoing works on Line 4 this summer to automate it as well

I've been on the metro in Toulouse as well, which is fully automated. Kind of cool because you can sit in the front where normally the cockpit would be. Toulouse is of course a major Airbus city.

There are many automated train systems around at different levels, more than we think. https://en.wikipedia.org/wiki/List_of_automated_train_system...

This is a highly underrated insight to autonomous system design. The use of autonomous systems in dense public spaces will usually be driven by programmatic constraints before technical ones.

It is very difficult to convert trains to run fully automated not because of the technology but because of unions that typically run the show. Don’t underestimate the power of the unions. They have decades of experience in getting their way.

Very US specific.

Why do you think that?

Plenty of examples in the rest of this thread are talking about automated lines in France, which has some of the strongest unions in the world. So the idea that unions are what's blocking automated systems is crazy, at least if applied globally.

I've always been amazed at the trams in Vegas between the casinos that are I think remotely semi-automated (at least one of them, I've seen an operator in a booth at one terminus, in the Excalibur). Some of them are even elevated! I think the Las Vegas monorail is also automated.

I rode the Las Vegas Monorail, which is automated. It is also elevated. When it gets to the end of the line, they tell everyone to exit the train. Well, I didn't (hey, there's no attendant to make me...)

They then switched from one track to the other. This involved moving down past the station to this stub of track that just ends in midair, with absolutely nothing except the control system to keep you from going too far and crashing down 15 or 20 feet to the ground below. That was... a bit nerve-wracking. It all worked, and they don't regularly fall off and crash, but... I don't think I'd do that again.

But yes, they're automated, and it works, and works reliably.

And if it works in one of the places in the world where people can carry open alcohol, it's pretty foolproof? :)

Love the part about not getting off the monorail. I'm tempted to try that myself someday!

There are still safety critical things that the Airbus is unable to do itself. For example talk to ATC, make decisions about changing weather conditions, fly a visual approach, or know that the causes of an electrical problem is coffee spilled on the thrust levers.

Exactly, there is a limit to the intelligence that people are willing to pay for. For example, no aircraft manufacturer will go as far as to develop a differential thrust system that replicates how UA 232 pilots guided their plane after flight controls are lost. Its just not cost effective.

Is it really any different from a self driving car thats designed to pull over to the side of the road safely if its main functionality fails for an unexpected reason?

Yeah, aircraft can't pull over to the side of the road in a catastrophic failure, and aircraft manufacturers will not pay to have the unique actions that a human faced with death would take built into the aircraft systems.

It's not necessary to reproduce all human abilities in recovering from emergencies because about half of crashes are caused by the human pilots in the first place. Just automating the routine things and letting it crash in exceptional cases could be safer than having a human that can save it in the exceptional cases but also mess up the routine things occasionally.

Although we have no idea how many potential crashes become near misses as a result of human intervention. For example a pilot initiating a go around in response to a runway incursion. I absolutely agree that most of those issues could be avoided with better technology. But that may require changes to all airports, all aircraft and perhaps even all airport vehicles. Suddenly every single approach is category iii, all the time without any alternative.

Why would there be coffee spilled on the thrust levers of an automated airplane? There is not a robot in there drinking coffee, and presumably they don't just put the thrust levers in the cabin with the coffee drinking passengers. You don't have people in the wheel wells either debugging issues, you rely on the systems to work.

AI can definitely talk to ATC, in fact it could disrupt ATC who could send declarative commands to planes to remove chance of miscommunication.

Even my Telsa makes decisions about changing weather conditions (eg. disables automated lane changes in heavy rain & such, AI controls the wipers)

The coffee is a reference to this incident -- https://ukaviation.news/spilt-coffee-caused-smoke-and-fumes-... -- in which coffee spilt by the pilot caused an electrical fault. Apparently there is quite a lot of high-current equipment in the avionics bay and it is not ingress-protected....

I love to know if we use https://www.youtube.com/results?search_query=air+crash+inves... (Air Crash Investigation Play List ) as test cases for the current Airbus system, how many crashes, or near misses can the automation system handle?

In Singapore, many MRT (Mass Rapid Transport, a.k.a Train) lines are fully automated, there are no drivers to drive them around.

Vancouver's SkyTrain (Elevated/underground commuter rail line) is 100% automated - and the first branch of it was built in 1985.

Unless they changed it in recent years, I believe it's control systems currently run on OS/2.

One of the few things a pilot needs to do (in normal operation) is to enter the take-off weight of the plane, but they get that wrong sometimes: https://australianaviation.com.au/2011/12/incorrect-data-ent...

Around 9/11 I was surprised the airplane manufacturers didn't put in a "big red button", that once mashed would irrevocably autoland to the nearest airport.

they've had the capability.

No they haven’t. Flying is the easy part. The hard parts are navigating and communicating. Even then, even the largest airports will only support precision approaches on some runways in some directions because maintaining ILS equipment is expensive. What happens when you push the button and there’s a massive crosswind across the ILS runway? Also communication standards have ossified such that the only way you can talk to ATC is through 50s era radios. That means any fully autonomous aircraft would have to be able to accept verbal commands and produce verbal acknowledgments. The technology is so bad that a simple stuck mic can shutdown an airport. https://youtu.be/Wcq6KipBOro

I was thinking more "broadcast we are under attack, or pilot unconscious" as the communication, and land (subject to external control?)

They've had autoland for a while. Maybe in this context it means the nearest ILS runway, or nearest externally directed ILS runway. (just not landing in the side of a skyscraper)

Yes and auto land capability is not always available. Even today object avoidance is a hard issue. What happens when the aircraft flies into a flock of birds or some other obstacle, like another aircraft on the way to the runway. What happens when two planes both push the button at the same time and decide to both use the same runway? Autopilot even in aviation is a misnomer.

What happens if they fly into a building because they don't have the button? It's a management of risks.

Besides, there can be a back-link and information updates for the landing box to manage situations with two or more planes in emergency mode.

Having a unpredictable button which may significantly increase risk to an aircraft and others is not risk management no matter how dire or remote it’s predicted usage would be. First thing any knowledgeable malicious actor would do is pull the breakers associated with the system or simply disable it. This button would have a way to disable it assuming accidental or malicious activation right?

You will have a hard time creating a situation where pushing the button for an emergency landing creates a disaster.

It's very easy to imagine: the plane lands on top of another plane at a busy airport.

That happens if everything goes wrong but you cannot create that situation intentionally. Usually, pushing the emergency button would send an emergency signal, and air traffic control would clear the airport. Landing on top of another plane is an emergency within the emergency - very, very unlikely.

You can't just "clear" a big airport. There's also no way to guarantee that the emergency signal would be received.

You also can't guarantee that you won't be hit by a meteorite and yet you leave your house.

If you worry about the signal not being received and the inability of air traffic control to spot the irregularity, then only initiate safety landing when the plane receives an answer-signal. Otherwise, let the plane fly however it wants and let it crash somewhere randomly because most places are better equipped to handle a random plane landing than a regular airport.

I think you're forgetting that the button itself would only be useful in extremely unlikely contingencies. But sure, you can spin a story about a bunch of interconnected automated systems always working correctly in highly unusual circumstances. Doesn't usually pan out that way in practice.

Autoland isn't possible on just any ILS. Autoland doesn't exit the runway either.

Correct, but at least there is "brake to vacate" the runway.

https://en.wikipedia.org/wiki/Brake_to_Vacate

When a pilot declares an emergency s/he can rely on the airport and airspace around it being cleared by flight control. If a hypothetical red button was pushed, there would be a standard procedure the aircraft computers execute which in turn can be relied on by flight control.

You can buy these for smaller airplanes: https://www.garmin.com/en-US/autonomi/

Basically a "pilot is incapacitated" button, it will pick the nearest airport and attempt to land there, and will even communicate with ATC. I believe the idea is that if you're a general aviation pilot flying with your family and you keel over, they can press this button to get down safely.

The problem is security. How do you prevent that such a system is unhackable and can't be misused remotely? That's the number one concern in all of these industries these days.

This, this is why we can't have nice things. I understand the hacking concerns, but really, it seems that we've constructed arbitrary and next-to-impossible problem-space boundaries that appear to only stifle change and solutions to big problems.

The same way you protect all other flight-essential electronics? You air-gap it.

Garmin ships a feature called autoland. I think at least one plane is shippng with that.

Austin Meyer (of x-plane fame) actually build an ipad app that can be hooked up to the autopilot (he actually did this at some point but never shipped the integrated product probably for certification reasons) to do exactly this. It's called xavion: http://xavion.com/. Basically the app calculates the safest glidepath to the best airport for you and guides you to that. The feature is designed for engine out scenarios where pilots don't have a lot of time for making good decisions. It basically continously updates its plan as you are flying based on glide ratio, altitude, runway length and a few other factors.

The autopilot integration is probably not all that hard to do. It's just tricky from a legal point of view.

I heard similar from a retired pilot in about 1994.

@PinguTS, well your friend is either mistaken, or you misheard him or a combination of both.

First, auto-takeoff is not possible. Its not possible due to technical constraints (e.g. concerns regarding LOC capture on the ground). It is also not possible due to safety constraints (e.g. the requirement to be able to deal with an aborted takeoff in a timely manner).

Second, automated landing IS used, but it is only used in limited situations. Why ?

- Because not all runways are equipped for CAT III autoland (i.e. the aircraft is not capable of landing itself without a shit-ton of calibrated equipment on the ground being there first, also good luck finding a CATIII autoland somewhere mountainous or with an otherwise tricky approach).

- Because not all aircraft are equipped for CATIII autoland (see above, more kit, more money).

- Because the autoland system itself has DH (Decision Height) and RVR (Runway Visual Range) requirements. The exception to this is CAT IIIC autoland, but AFAIK the vast majority of CATIII certified runways are only to CAT IIIB because that's expensive enough to certify, even for major airports.

- Because of general weather requirements. For example, I don't know about you, but I'd rather NEVER see an aircraft attempting autoland in a Captain's crosswind (i.e. crosswind so strong that the airline prohibits the First Officer from flying it).

- Becuase autoland operations at airports involve ATC increasing traffic separation. Increase separation leads to what ? Yes, that's right... decreased flow rate and increased possibility of delays and cancellations.

> Because not all runways are equipped for CAT III autoland (i.e. the aircraft is not capable of landing itself without a shit-ton of calibrated equipment on the ground being there first, also good luck finding a CATIII autoland somewhere mountainous or with an otherwise tricky approach).

I think if you get must of the major airports covered that's still good enough. But, if I understood this article correctly, what Airbus is trying now is to make it happen without those ground requirements. And I'm sure they do not expect to cover every airport and runway to begin with.

I don't think the major airports would be good enough. What about emergencies where an aircraft has to divert? Can the automated plane land itself without engines, side slip to lose altitude faster and land safely on a non-ILS airport like the Gimli glider?

Will it do the same on a grass strip, which some pilot did without engines in the Houston area?

Can it do a perfect belly landing like a Polish pilot did some years ago?

Can it land in the Hudson?

All of those pilots had massive experience beyond button pushing auto-piloted aircraft.

I think (hope) this is just a typical marketing program which is not actually going to happen.

Maybe the automated plane cannot do all those things, but on the other hand it will not dive itself into a mountain because it is feeling very sad.

It will happily do that if the pitot tubes are clogged or other sensory input is wrong. Or if there is an MCAS situation.

Moreover, it is easy to find at least 30 successful landings under the above conditions, while the situation you are referring to has about 4 cases, 2 of them contested.

Also, locking the cockpit doors on airliners is stupid in my opinion.

Many crashes are also due to pilot mistakes. Or due to some past damages done to the plane by previous pilots.

Aviation industry follows safety in redundancy principles. So on top of all the redundancy they are going to add in a system like this, I assume pilot will be part of the redundancy too.

Perhaps they can remove co-pilot soon. But, it will take a few years to go without any pilot. Even then perhaps they start with cargo airlines. I assume they can use a ground station as an emergency pilot like how military drones work.

That's an insight. You use automated stuff only, when the automated stuff is available? Wow. Haven't thought about that.

Of course we are talking only about situations when its is available and some airline then still insists that you don't use it.

> That's an insight. You use automated stuff only, when the automated stuff is available? Wow. Haven't thought about that.

You didn't understand what he said. What you wrote is not an insight. The actual insight is, in air travel, you use automated stuff only when the automated stuff is available everywhere around you. Because your plane might have to divert due to an emergency and land at another airport/field that doesn't have any automation in place.

RNAV is making CAT ILS obsolete and might do autoland "soon" https://www.airliners.net/forum/viewtopic.php?t=1395929 (more detailed discussions)

Do you mean the “Airbus demonstrates first fully automatic vision-based take-off” type of not possible?

https://www.airbus.com/newsroom/press-releases/en/2020/01/ai...

My predominant aim was countering the wholly incorrect "automatic take-off was possible 15 years ago but the airlines don't use it" statement put forward by the OP.

However, since you ask, I would say two things : (a) The very Press Release you link to does take care to manage expectations (e.g. "Airbus’ mission is not to move ahead with autonomy as a target in itself").

(b) Your automatic take-off in 100% of situations is never going to happen. Sure, on a perfect dry day with nil-winds you can conjure up a great sounding Press Release. But throw in some "real world" and you can forget it. Cross winds, monsoon flooded runways, engine failures, bird strikes, failures of electronic systems (yes, it happens, even with triplicate redundancy) ... the highly trained humans at the front are not going anywhere, nor is their training regime going to be cut back.

Right now unsupervised learning techniques aren't able to teach a meaningful world model to a neural network.

But this won't last (hopefully), thanks to our brilliant co-humans doing cutting-edge research on that subject. And when a neural network is able to differentiate and anticipate possibles actions of winds, engines, birds and onboard electronics, then you have an autonomous take-off on par with humans pilots, enhanced with an instant feedback loop with built-in checklists and avionics.

It's a big stretch of course. But never say never, they say.

> First, auto-takeoff is not possible

So was auto-driving; engineering visionaries don't stop at what is currently possible.

The problem with automated air transport is that the failure mode is going down through gravity and probably not being able to land on a suitable surface. I doubt we have automated pilots developed so far that any failure (and those are a lot!!) can be handled, and even then you might want to manually control.

The failure mode of automated driving is less severe. You're already on a road, gravity holds you down and unless the brakes don't work (although in an electric car or bus, when not accelerating you're braking) the likelihood of accidents is smaller.

It’s always possible to have a remote control station with extremely well trained emergency pilots to take over in those case. That of course requires a very reliable remote control system.

My first job was based at Cranfield University which had an attached airfield - and back in the 80's they where practicing auto landings then

Link to Airbus's press release: https://www.airbus.com/newsroom/press-releases/en/2020/06/ai... More info here: https://www.airbus.com/newsroom/stories/autonomy-aerial-mobi...

Airbus has an "innovation" (R&D) department and they're looking into more ways to have computers do the routine parts of flight.

Airbus's materials (linked) do mention "self-piloting", but I can't imagine that they are thinking of the scale as BusinessInsider is when they write "self-flying". No one's proposing getting rid of pilots, having a computer fly a plane completely would be insanity.

Would it be more insane than a self-driving car? Because to me, who knows nearly nothing about planes, it seems like it would be significantly easier to make a self-flying plane than a self-driving car. But obviously if something goes wrong the plane crash would be much worse than the car crash.

Drivers take almost all information they need for driving from looking through the windows, whereas pilots take almost all information from looking at the dials. That difference alone makes flying inherently more automatable. Automated driving adds a lot of sensing requirements over manual driving, but in flying the manual way already requires most of the sensing you'd need for automation.

The act of controlling a plane can be done entirely by looking at gauges, but that's not the only thing a pilot does. Pilots are also interfacing with ATC, flight attendants, dispatch, maintenance, and passengers. Even if you replace these interfaces with a computer usable format, you still need someone with judgment and responsibility for the flight. Pilots exercise their judgment over the information they receive from these other sources, and to coordinate these resources for efficient, safe flight. It's a huge challenge to replace all of this with a system.

I don't disagree at all. If you automated all the controls you'd still have some form of plane manager.

Perhaps a far more meaningful automation push could be done by creating an entirely new ATC scheme with automation in the air and on the ground, designed not defuse the class of mistakes that are typical for humans but to work around weaknesses of automation while exploiting its strengths (relative to humans). I suspect that the error margins needed for one group have very little overlap with the error margins needed for the other and with the general multilevel architecture of existing ATC it should be possible to designate a subset of airspace to the new system and having then coexist.

> whereas pilots take almost all information from looking at the dials.

Under normal operation, yes, a lot of information is available there however all pilot training is underpinned by learning to fly the plane by eye, without relying on avionics. Mainly due to the fact that they can and do fail in new and exciting ways.

Even a plane's interpretation of its own configuration can be incorrect, resulting in automated systems believing a given control plane is oriented a certain way, when the reality is very different.

Depends on what you mean by self-flying, which is why I pointed out that BI probably has different ideas than Airbus.

To most people (and BI?) it will mean "taxi, takeoff, navigating/obstacle avoidance, landing". To the extent that driving a car isn't "what you see is what you get", flight is much worse.

In the simplest case of flight, you are in cruise (neither takeoff nor landing), and depending on what kind of airspace you're in you may need to: * listen to ATC for commands, and advise them of what you're doing, * avoid weather, * look out for traffic, * watch your gauges, * stay on a heading, altitude, and generally fly the plane. (This part is routinely done with autopilot.)

In short, if it seems easier, it's because pilots make it look so casual.

A thing they teach pilots is: "Aviate, Navigate, Communicate". This is the basic priorities for your attention. If you're pointed at terrain your priority is to fix that first. The point of autopilot today is to free you (to some extent) of that first responsibility, especially during routine parts of flight, so that you can do the other things. Airbus is looking for more things to automate to let pilots do more strategic thinking in general.

Contrast this with a car, which is probably more in line with what BI is thinking. The kinetic energies are much lower, there's no need to communicate, the rules for driving are pretty straightforward (get to where you want to be, without hitting other people, while respecting traffic lights and speed limits), and to top it all off, if the AI feels that it's in over its head than the car can just stop whenever and everyone can climb out of the vehicle.

A car can't just stop in the middle of an intersection. Turn signals and traffic lights are similar to air traffic control. The risk of collision is much higher in a car than a plane, this is why you see many more traffic accidents than plane crashes. In both cases the rules are get where you want to be without colliding with anything or making passengers uncomfortable.

All good points, but just a side note on one of my reservations about self-driving cars:

> and to top it all off, if the AI feels that it's in over its head than the car can just stop whenever and everyone can climb out of the vehicle

Just stopping in a car is always less dangerous than just stopping a plane, but there are cases where just stopping is very dangerous. Desert driving in the absence of sandstorms is probably unlikely to overwhelm an AI. However, in Alaska or the upper Midwest, you'll get snowstorms where you've got 6 inches or more of snow on top of the road, IR and the visible spectrum are limited to tens of feet due to blowing snow, exposed flesh will begin to freeze in under 10 minutes, and you might get hypothermia if you're in your car overnight. Getting out of the car and trying to flag down a passerby is likely to get you struck by an out-of-control car, so your best bet is to call a tow truck and hope they can get to you before the inside of your car gets too cold. In major blizzards, there are so many cars needing rescue, that some people need to wait in their cars until well into the following morning. The local newscast will warn people that driving is potentially life-threatening, but that doesn't mean that essential workers won't be driving. Having half of your trauma surgeons stuck in ditches in self-driving cars during a blizzard isn't a great situation to be in.

People keep candles, matches, and candy/energy bars in old coffee cans, plus water and blankets in their trunks (boots, if you're British) for these circumstances in order to be able to shelter-in-place for a day. Even humans with years of driving experience under icy conditions are pretty bad at driving in these conditions.

Having grown up in the upper Midwest, I was amazed how terrible drivers in the D.C. metro area get with just a dusting of snow or slightly icy conditions. I'm a bit worried that California-based automated driving companies will miss lots of common corner-cases associated with Winter driving in Alaska / the upper Midwest. I expect that if the road is uniformly icy, an AI would adapt fine with the assumption that traction levels here are the same as those up ahead. However, due to lower thermal mass, bridges undergo larger temperature swings and are likely to be icy. There are a bunch of gotchas associated with driving near active snowplows. Sometimes it's best to drive in the ruts created by other cars, but sometimes that's where the road is iciest and it's better to drive offset from the center of your lane. Under normal conditions, your car has a much shorter stopping distance than a loaded tractor-trailer, but in some icy conditions, the much higher tire loading will allow the tractor-trailer to stop in a shorter distance than a passenger car. Many people put sand bags in their trunks (boots) to increase their tire loading during winter.

There are just tons of little rules like this that you're either going to need to hard-code, or get good simulations plus a lot of driving time in International Falls, and hoping that a closed track still gets you conditions that teach the AI when to drive in the center of the the lane and when to drive off-center. Hopefully the AI also learns to visually identify different kinds of icy or likely icy patches.

I'm sure it'll all get worked out, but I wouldn't be surprised if the first Winter or two with self-driving cars ends up with a bunch of cars stuck in ditches in Alaska and the upper Midwest. It might look a bit like the newscasts from when they get icy roads in Texas or Georgia for the first time in decades.

Getting back to aircraft, even for everything that's in the emergency procedures manuals, writing good realistic simulations in order to get enough training data for the AI is going to be very difficult. For situations that aren't in the emergency procedures manuals, both most humans and most AIs are going to be in deep trouble.

The "bail" maneuver is considerably easier in a car.

It's also far more likely to be necessary, but pulling over to the side of the road is easier than landing off runway.

"Airbus has an "innovation" (R&D) department"

They have pretty deep partnerships with European state schools etc. which is highly rational, at the same time, it puts issues of 'government subsidy' on the table re: trade agreements. Nobody makes such complicated things 'alone', at that level it's going to be a broad mix of national integration efforts. It makes it hard to do trade because everyone else is pointing out everyone else's non-market subsidisations.

this is true of most if not all companies in the knowledge sector. Google for example is known for starting as a phd project.

It's not remotely the same with G as it is for AirBus.

Commercial Airlines do not exist without government intervention, and often depend on non-market entities for key technology.

Google could drop every academic program and not skip a beat, as most companies.

Maybe one day we'll have pilots reserved for emergencies, but only on the ground like drone operators. Hotswapping between flights that need corrections as problems occur.

Obviously not going to happen tomorrow and needs more robust communications infrastructure over the deadzones of the planet but it's fun to imagine. I trust machines more than people, unless they're made by Boeing.

I would imagine something like Starlink could help with the deadzone issue?

We already have satellite communications for media and voice, but iirc they have fairly limited bandwidth constraints. I think figuring out a way to get the flight recorder data to operators in realtime would be the first problem to solve, and I don't think it's been done yet for commercial flights.

There was a startup around like 5-6 years ago (gone now, can't find any info on them, their name was "Flamingo" something or other) that had a proof of concept solution but they burned out quickly.

Latency is probably the bigger issue for aircraft, but Starlink could help there as well.

Can Starlink work with a vehicle that is in motion? And at 500 mph? I had assumed that it only worked with a stationary transceiver.

Perhaps not initially, but IIRC, they use "digital" antenna so it should be upgradable to track in real-time. Perhaps even by an OTA update.

There is nothing stopping the use of a traditional auto tracking antenna either.

Except for accuracy, mechanical wear, price, and size.

Price is in five figures, and the others are not real problems. How do you think datalinks on military aircraft work?

Yes, it can, although I can't be sure it will work at 500 mph. They tested it with a C-12, which has a top speed of 338 mph, and were able to transfer data at 610 Mbps.

Source: https://spacenews.com/spacex-plans-to-start-offering-starlin...

Yes. The USAF has already tested it successfully on several aircraft.

A Starlink receiver always has the satellite(s) in fairly fast motion, and phased array antennas can "move" quite quickly, so it's possible that there's no particular challenge to an aircraft scenario. Then again, the basic rooftop unit may make some simplifying assumptions and aircraft would require better hardware and/or software, but that's not a big problem as they can handle a higher price point.

Imagine someone piloting a plane with at least several hundreds of milliseconds of lag.

What do you think is better about people made by Boeing?

If ever. Taxi ride costs $50 per hour while flight costs $50k per hour.

Getting rid of driver in a taxi saves 50%, while in plane saves 1%.

There are a lot more taxi drivers around world (although I’d guess they generate similar or just slightly higher value).

Automation can (eventually) significantly reduce pilot error; it also has a more reliable way of fixing corrects and avoiding regressions. Automation isn't just about the cost of the person flying the plane, at some point in the development curve it is just a better choice (e.g. why factory robots can do better work than manual assembly workers).

I agree with that. What would be interesting to read is how is data collected right now? Is it compared between automation and manual operation. Who is actually doing data crunching - airlines, Airbus or avionics supplier?

From what I've read it seems that the "leader" Tesla's fleet learning is largely a myth.

I’d love to read more about Tesla’s fleet learning being a myth

The FAA is really serious about making sure accidents don’t happen for the same reason twice. That involves changes to hardware as well as pilot training, they crunch the data for the USA, similar authorities do it for other countries as well.

Unless you get Boeing 737 MAX where pilots fight the automation and lose.

1% extra profit is something competitive airlines like Ryanair would no doubt appreciate. Also it could ensure the aircraft operates at optimal fuel efficiency, which is central to their business model. Maybe for national airliners these marginal gains matter less and automation could be a bad look for them.

I wouldn’t trust a pilot to take over from the computer during a crisis if they’ve barely done any real flying in the past few years. Humans need practice to become and remain skilled. I’m worried about the effect of such systems on pilot skill.

There's some interesting articles about the role of automation in the Air France 447 disaster:

Short article:

https://hbr.org/2017/09/the-tragic-crash-of-flight-af447-sho...

Much longer article, but 100% worth the read if you have time:

https://www.vanityfair.com/news/business/2014/10/air-france-...

Right.

This is a big issue for self-flying aircraft. Keeping the aircraft stable requires reliable altitude, airspeed, and orientation data. If you have that, you can stay in the air. AF 447 lost airspeed data, the flight control system dropped to a lower level, and gave up stall protection. It couldn't tell if the aircraft was stalling. That was enough to confuse the pilots so badly that they crashed a working airplane with plenty of speed, power, and altitude into the ocean.

A self-flying aircraft has the same problem, only worse. Without that info, it cannot stay in the air. There are fighters that need so much active stabilization that if you lose air data, you have to eject. Self-flying aircraft thus need really, really reliable air data sensors, and lots of them.

Getting airspeed from GPS has been suggested. There's one bizjet which does this. It can't be flown in areas where GPS jamming is in progress. Over-dependence on GPS is a problem.

All that is just to be able to fly straight and level. That's flight control. Navigation and landing are separate problems.

All this and the code needs to be written correctly especially with regards to edge cases. Thousands of planes have been flying around the world for several decades now and we still see unique accidents or slip-ups that have never occurred before.

Think of all the edge cases that need to be accounted for with all the different variables out there from weather to people misbehaving on the plane. It's only going to take one freak accident to have people not want to board a self-flying craft ever again.

Self flying doesn't mean no security guards or flight attendants, nor does it mean the plane can't make emergency landings

The only practical way to mitigate this that i can think of is more emphasis on regular simulator training. It may ultimately undercut the financial benefits of full time autopilot if pilots have to spend a larger fraction of their time on the clock in simulators, plus more sims would have to be built and operated.

The cockpit could just double as a simulator in non-emergency situations.

simulators might be good enough to train for most situations, maybe not to the same degree as if the pilot had been flying the whole time.

The situations that call for a pilot rather than an AI can only be trained for in a simulator.

The (few) hours in a simulator are probably far more valuable in an emergency than the thousands of hours of real flight.

(sport pilot here) Can you explain what are you thinking about? I had a few cases of unexpected problems while flying, simulators in my case would not help at all, but it may be different for large planes with huge inertia vs my larger-than-a-kite extremely sensible 2-seater.

You can't induce emergencies in actual passenger aircraft for the purpose of training. So your only option is a good simulator.

"Russian jet crashed 'because captain couldn't land without autopilot"

https://metro.co.uk/2019/05/16/russian-jet-crashed-captain-c...

The jet crashed because lightning hit it, the plane did not have an feature to drop fuel and landed over max landing weight and there are other reasons.

Metro is not the most reliable source of information.

"N crashes prevented due to automation" - A headline that will/can never be written. Net result is that automation saves lives.

It's not as if a computer didn't crashe 2 737-MAX8 on take-off, despite pilots trying to correct the trajectory !

No amount of engineering genius can make up for human greed. Boeing took a gamble and they lost.

I wonder how much practice pilots get nowadays, when there are so few passenger flights worldwide.

Old pilot joke: "The airplanes of the future will have a pilot and a dog in the cockpit. The pilot's job is to feed the dog and the dog's job is to bite the pilot if he touches anything."

I'm all for this for the shipping industry and completely unmanned, but I'm not about to fly on this myself.

The article mentions that.

“For its passenger jets, though, Airbus states the tech won't replace pilots in the cockpit but will make flying safer by helping reduce workload. … "For autonomous technologies to improve flight operations and overall aircraft performance, pilots will remain at the heart of operations," Airbus said in a press release.”

Is that logical?

What are the environmental and economic costs of an automated cargo ship going down vs airliner (cargo for comparison)?

It’s hundreds of millions dollars and fuel.

Actually this leads to an interesting point..as far as I know cargo ships are not automated sooner. It seems that it's just as open as the sky...potentially the sea is much less forgiving?

I would mainly say less standardized and less regulated. Big vessels already require pilots in and out of ports because the crew isn't trusted. Then essentially every single ship is a custom build with all the years of problems to iron out that comes with.

With less regulated you don't have a required AIS on every canoe or vessel, meaning you need to react on visual and radar input according to the colregs. In the US every aircraft flying already have ADS-B.

So with some worldwide regulation and a modular system able to interface and work with different equipment and sizes to create the economics of scale you have something working in the not too far future.

Though, this still requires the bridge and interfacing systems and sensors to be stable enough. Meaning no vessel today is good enough.

Compared to aviation it is much more hands on based on tribal knowledge. Like on a ship I work on "the remote for the autopilot tends to quit working every 3 months, just reset the whole autopilot, by pulling the breaker, if it does." This information stays on the boat and never reaches the manufacturer, instead of fixing what is probably some overflow happening. And the fix would be replacing the unit instead of deeper troubleshooting. Seen it happen many times.

If you ever take commercial flights you are already being flown by autopilot, and have been for decades. It might give you comfort that there is a human pilot in the cockpit for backup, but it's only a matter of time before the human backup moves to a ground station.

The Boeing 737 max 8 software couldn't keep a plane in the sky with an army of pilots fighting to save their own lives.

I wouldn't get in one of these until there are better controls on this kind of software, it's not the same as autopilot.

There is a long list of entirely preventable human-caused accidents. Is there a reason pilot-caused crashes are less scary for you? Computer caused accidents will be fixed and won't happen again. Human-caused accidents will keep happening as long as experience is valuable.

    Aeroflot Flight 593 - pilot let his son fly the plane, 63 dead
    Germanwings Flight 9525 - (possibly suicidal) pilot deliberately crashed , 144 dead
    Air France Flight 447 - pilot caused airplane to stall, 228 dead
    Aero Flight 311 - both pilots got drunk, 25 dead

and this is just a random selection, there are long long lists of human-caused aviation accidents.

I'm not a luddite, if the software is ready and safer than people then I'd be okay with it.

There's a history of software not being ready while people pretend it is and then it kills people (Therac-25).

I'm just skeptical that we'll know when it's actually safe.

Therac-25 wasn’t a “it’s not ready yet!”-type issue. It wasn’t an expected or anticipated failure-mode - it only became a (literal) textbook case-study after people died and the industry has learned and improved as a consequence.

They ignored repeated failures and evidence of malfunction by saying it was “impossible” that it could be failing in that way.

Unexpected failure modes are the issue. The Boeing 737 max 8 failure being tied to one sensor would suggest the industry has not fully learned the lesson.

My understanding is that it was a UX issue - the "malfunctioning" was the system working as-directed by the user, but the UX was horrible for informing the user what they were doing.

It was a lot worse than that: http://sunnyday.mit.edu/papers/therac.pdf

That paper is long, but does a great job of giving the context and going deep on the details.

Thank you!

Don't forget Colgan Air Flight 3407 where the pilot was either too sleepy, or simply didn't know what he was doing.

Most accidents are due to human error, so you are taking way more chances by relying on the ability of pilots, statistics wise.

Doesn't this just mean that today the mechanical parts of a plane are pretty reliable?

It's not much of a comparison to flying software that isn't yet in use since there isn't any data to make that comparison.

Am I wrong? Are planes flown today without human intervention?

Apart from landing and take-off, most flights are in autopilot for most of the flight duration nowadays.

The 737 max was brought down by a classic autopilot style feature.

I imagine full automation will come to air cargo first and gradually find its way in passenger liners as the public becomes less hysterical.

That isn't really true. It was brought down by stall prevention software that was using input from a single faulty sensor, and there was no way to override the inputs from this software. Further, there were multiple incidents before boeing admitted what was happening, even though in retrospect it looks like they knew what was happening all along.

My point was that it functioned in a manner vastly more similar to a conventional autopilot than what Airbus is proposing to do in this project.

MCAS was a simple algorithm that altered flight controls in a predetermined way upon a limited set of inputs. Airbus is proposing a vastly more ambitious solution that includes additional inputs from computer vision and a global view of the state of the aircraft.

Doesn't that make things worse?

If a relatively simple algorithm was not safe because of bad engineering decisions (or bad management incentives, whatever the case is) - then wouldn't a much more complex system be even more likely to have hard to discover corner cases and failures?

In this instance, the simplicity of the system was its down fall.

I think the use of human pilots complicates a system. You are relying on a component to the system that is susceptible to tiredness, distraction, threats, rage, revenge, self destruction, and sudden death. Complete automation would replace one extremely complex and unpredictable component, with a less complex and more predictable component.

The autopilot is still at the control of the pilots, and usually enabled only at higher altitude. Landing/takeoff are still manually flown by pilots most of the time.

I don't have issues with a computers ability to maintain altitude, climb, or turn to a heading. I have a problem with a computer's ability to respond to the unexpected while in the air. For instance, comms failure is a scenario pilots train for and can deal with. I imagine autopilot might have some issues with that.

I've read so many stories about drunk pilots, unlicensed pilots (primarily in third world countries), sleepy pilots, unskilled pilots (remember that aircraft that crashed because the captain was pushing down while the copilot was pulling up?), etc. that I think I'd be more comfortable flying with an AI.

The best landing I ever had was on a Boeing 747(?) coming into LAX probably somewhere around 2000.

We were flying through a really big storm front forever. Drop. Rise. Drop. Rise. For hours.

Finally, we start coming down still bouncing all over the place. About 30 minutes outside LAX, the plane suddenly got as smooth as glass and we touched down with the lightest touch I had ever experienced.

Captain comes on: "Well, we were planning on diverting, but you can all thank the fine engineers at Boeing for bringing us down in that weather on autoland."

Guy next to me who threw up: "Why the hell couldn't they have turned that on 2 hours earlier?"

Pilots are done.

The shipping industry is one where it could be extremely helpful, particularly in thwarting pirates. I've always wondered why cargo ships are not drones, full autonomous until it gets within a certain range of its destination.

I'm not sure you'd like the fully automatic sentry miniguns associated with drone ships !

Classic NIMBY'ism :-)

I thought most of plane take off, flight, and landing were automated long ago... ?

[sorry, should have done more googling: https://www.flightdeckfriend.com/can-a-plane-land-automatica... seems to say auto-land is there (but used very little) and no auto-takeoff]

It has been, but there is still a strong incentive to have a human in semi-control for contingencies

Not only that, but it's free training. This helps keep pilots sharp and current.

If autoland fails it could be that there's something else going wrong with the aircraft(ranging from a faulty sensor to something more serious). Now you have a pilot with degraded proficiency trying to land in an unusual condition. This impairs safety.

I believe the A320 has some auto land capability. Not a pilot, so don't know the details, but based on Youtube, it appears the pilot has to do some configuration on approach (line up the plane to some extent, set engines appropriately, etc).

Yep, https://en.wikipedia.org/wiki/Instrument_landing_system

It does require hardware on the runway though.

This is for landing without visual contact with the runway (i.e. instrument flight conditions), typically the pilot is still controlling the aircraft, and looking at an instrument that tells him/her how to adjust the controls, although many aircraft can land automatically using these same systems.

There are other people talking about autoland, but it's actually meant to line the aircraft up with the runway and gently glide it down, and then the pilot deactivates it around 200-300ft off the ground to flare. It's not meant to perform the entire landing sequence autonomously. Autoland doesn't flare by itself, so while it can safely land the aircraft, it will cause a ton of stress on the landing gear that adds up to more frequent downtime for repairs.

Now that's just completely wrong. Airliners had full automatic landing capability for a long time, it's very often used when there's zero visibility at the destination airport.

Maybe FedEx, UPS, and DHL should start first?

Possibly they could go to single pilot operation on freighters?

or Amazon with their prime fleet, I see then more willing to test this technology since they love cutting cost

I would not be surprised to see self-flying planes take off (pun intended) before self-driving cars.

It's a much easier problem, which is why planes have been mostly automated for a while now.

The equivalent for cars would be if we only allowed them on almost entirely empty freeways and highways, and on designated, fairly uniform and well marked parking lots directly off those freeways. Self driving cars can already do all that fairly well. It's all the fiddly bits that planes don't have to deal with (pedestrians, city roads, weird/bad roads and signage, lots of other cars, etc) that make self driving cars hard.

Don't forget planes all file flight plans ahead of time so not only is there more space, but it is relatively easy to check for potential conflicts before flights leave the ground.

Flight plans are filed in advance, but a flight plan is just a plan. It's an impossible task to predict where an airplane will fly in advance, there are so many variables involved.

Just some of the many reasons a flight will vary from flight plan:

-Avoiding convective build up (thunderstorm)

-Avoiding turbulence for passenger comfort (changing altitude)

-Unexpectedly high headwinds (changing altitude)

-A cold front causes winds to shift, so the runways in use change

-Low clouds limits arrival/departure rate, so holding is needed

-Snow removal from runways/taxiways

We have these with Predator drones etc. These were mature at the time when the DARPA Grand Challenge had robot cars crashing in the desert.

Predator drones aren't exactly self-flying though, they are remotely piloted.

They self fly like the more expensive quadcopters do. You can set GPS waypoints and they will guide themselves through the route. Then you can click a button and they will orbit a defined point in the sky. The operators are usually just working the sensor pod because that's the actual hard and important work

Part of me wonders, in the future, how we'll view the attempts to automate pilots and drivers.

It feels like a natural step now in the technology but perhaps it'll be viewed like trying to make a robot ride a horse instead of looking to invent the car.

I wonder what the next order-of-magnitude leap for transport will be. Cheap, fast, safe. Hyperloop maybe?

Will we look back at holding on to air and road travel as a bit of a reckless endeavour?

If you took humans off the road and made the roads suitable for only automated cars it would be so much easier to automate, its only hard because we have to make it drive with the signals/lines and such that humans require rather than just broadcasting things like the traffic signal in an easy way for cars to understand.

And when there's zero visibility?

ILS presumably - https://en.wikipedia.org/wiki/Instrument_landing_system

So would it not make sense to tie the two systems together in an intelligent way?

> So would it not make sense to tie the two systems together in an intelligent way?

There's no need to do so since ILS autoland has been a thing for decades and can be used by many aircraft today (from both Boeing and Airbus)

All the 'self-flying plane' has to do is engage the Cat III approach if it is available and let the old and boring autopilot do its thing.

Alternatively, if Cat III approach is unavailable, it can do what pilots already do: divert to an alternate airport.

https://www.flightdeckfriend.com/can-a-plane-land-automatica...

In Cat 3 autoland the plane does the approach and landing manoeuvres but the pilot is still in control. Complete zero visibility landing is not allowed. Pilot needs to see something like 50 feet to see the center line and make sure that the plane is in touchdown area. There are also wind limits, almost no turbulence is not allowed and so on.

That's not entirely true. Cat IIIc is "zero zero" autoland capable with no decision height required.

Does anyone think that if we go to ground-control backup pilots, they'll be monitoring second-by-second as if they're actually flying, and able to save a situation like US 1549? Or even other less dramatic urgent situations? At that point, why not keep them in the plane?

If you go to ground control monitoring -- the only point of it being that pilots could monitor multiple flights -- you are saying that the pilots are there to fix longer term issues (like flight planning, diversions), etc. Things that don't happen on a few second's timescale.

That might be acceptable on freight/cargo where losing a plane now and then is a risk you can deal with. Not with passengers.

I don't think anyone is willing to go that far.

Are there any real cost-savings to pilotless planes though? You’d still have a cabin-crew, too.

You probably don’t need a crew for freight only flights.

You save the cost of the pilots, and the cockpit (which would reduce the aircraft price and weight significantly, and increase capacity).

Might be hard to imagine for you now, but could be completely different story in a few years. Basically there’s no reason to believe people will be always more reliable than machines.

Mildly terrifying when we know how software is done at massive corporations like Boeing and Airbus : /

You could have stopped at "done".

> Boeing and Airbus

I wouldn’t even put Airbus in the same sentence with Boeing; they are using formal methods since 2001[0], whereas Boeing has outsourced its software to $9-an-hour engineers[1].

[0]: https://www.di.ens.fr/~delmas/papers/fm09.pdf

[1]: https://www.bloomberg.com/news/articles/2019-06-28/boeing-s-...

How bad does it have to get before the automated systems refuse to deal and a human has to land the plane? An engine gone? Two engines gone? Multiple engines gone and a really stiff crosswind? How good is it at landing in a blizzard on an icy runway? I know it never happens. Planes never land on the Hudson River, either. There's obvious value in making the happy path as safe as possible, and I'm certain full automation is the way to do that, but what are the current limits of automation?

Probably the limit is that in the miracle on the Hudson, Sully wouldn’t have had a co-pilot.

I suspect that the with the current electrical system on airbuses, the emergency electrical configuration wouldn't have enough power to run all those image processing computers at the very least (dual engine failure would cause this). However that could be resolved with additional battery capacity.

A more likely cause of automation failure would be failure or disagreement in the various sensors on the plane. These are typically triple redundant, but there have been cases where, for example, all the pitot tubes ice over and airspeed indications are lost. There are ways to deal with this but they're not currently programmed into the autopilot.

Other possible causes include failure of major flight controls requiring workarounds - eg, dealing with a stuck rudder with deliberate asymmetric thrust.

That being said flight control issues can probably be dealt with, if necessary. The real problem is with dealing with ambiguous situations where you have to weigh risks.

Consider: your radios are out. Per standard procedure you should proceed on your flight plan to the final fix and hold until your scheduled arrival time. However, fuel consumption is high - you're not sure, but you might have a slow fuel leak, and your destination has dicey weather. Do you divert without clearance or risk that you might be marginal on fuel at your scheduled arrival time?

Today's autopilots are designed to be as simple as possible.

The intention is that a pilot should be able to fully understand the complete autopilot algorithm and understand exactly how it should act in any situation.

Today's autopilot can't handle any emergencies at all, they are designed to disconnect when they detect something weird. They don't even understand aerodynamics.

I don't know where Airbus' new system lies, but if we actually designed a fully autonomous plane using modern traditional AI techniques (so no machine learning or neural networks; Just decision trees, adaptive aerodynamic models and predictions) then it could handle a wide range of emergencies.

It should be able to handle bad weather and engine out emergencies just fine, and should even be able to land in the Hudson River or anywhere else it has detailed maps.

These are all documented flight emergencies that any programmer of a fully autonomous plane would know about and test their software against in simulations.

The main areas where such fully autonomous planes would have issues are avionics failures, misleading sensors, emergency landings in areas with substandard mapping and unexpected aerodynamic failures (beyond what their adaptive aerodynamic model can handle)

I wonder how a computer would have reacted to Search US Airways Flight 1549...

Whats the publicly available data on the crash rate of autonomous drones in the military, unrelated to targeted attacks. It is the best indication, noting that they fly in adverse weather and uncertain routes with possibly less visibility to fall-back systems, and on long-delay operators.

I would expect domestic autonomous flying to incur at least comparable rates of failure to the military for routine operations with root causes in weather, or systems problems, given the likely sharing of weather, and IT systems behind autonomous vehicles.

It would be interesting to see take off, cruise and landing distinctly to understand their risk profiles.

There is also the public perception problem: We tolerate a level of crashes in air transport, predicated on a belief the system is self-correcting. the 737 MAX was withdrawn from service. The lithium battery problem in the 787 was identified. The square windows in the Comet were identified as a root cause.

Will we tolerate a 0.0000001% effective crash risk rate, from autonomous vehicles with 400+ passengers?

What's the error rate of pilots compared to computers?

If we, today, completely eliminated all human pilots for all flights on all aircraft and replaced them with the best automated systems available in one fell swoop would deaths increase or decrease?

I'm sure there would be incidents where "A simple software bug killed 200 people today".

I'm also sure that bug can be fixed, and there are also "dumb human error killed 200 people today".

I don't know if any amount of human skill, trained for an entire human lifetime, using the best pedagogical methods, with infinite resources devoted to training, can beat a computer in the long run.

Navy jets have auto-takeoff, tons of passenger planes have auto landing (which, apparently isn't as smooth as human landing), and planes spend most of their time on autopilot already. I think some military UAVs can do all or most of this stuff already. What are their accident rates compared to commercial airlines?

This is just a deathly accident waiting to happen.

We don't have the technology to make airplanes or cars respond to truly unexpected conditions. We have instead a series of (very) sophisticated one-trick ponies, which isn't nearly the same thing than an intelligent system.

Whoever dubbed ML "AI" was a marketing genius.

So was the person who dubbed statistics ML

from edmundhuber:

> Airbus's materials (linked) do mention "self-piloting", but I can't imagine that they are thinking of the scale as BusinessInsider is when they write "self-flying". No one's proposing getting rid of pilots, having a computer fly a plane completely would be insanity.

It's not incorrect to say that automated technologies aren't always better than human operators. But that's not the measure we need to use. The question is does the net failure rate increase or decrease? Say the automated technology fails 2x as much in unexpected failure cases, but halves typical failure rates?

Say an airline experiences 100 crashes a year due to unexpected failures, but 1,000 crashes a year due to expected failures (immense crash rates for any real airline, this is just an analogy). Using an automated technology might increase crashes from unexpected failures to 200, but reduce crashes due to expected failures down to 500. This is still a net decrease of 400 crashes. Assuming other things are equal, namely the severity of crashes, this is good outcome.

Of course in practice, we know that people's fears are not rational. Things like shark attacks and nuclear meltdowns scare people in a manner vastly disproportionate from their actual impact. It could be that the public has a lesser reaction to a plane going down with a person at the helm than an automated flight crashing.

Also, I'm not sure how ML or AI relates to this. It's not mentioned in the article, and I'd be surprised if any part of this system makes use of machine learning or AI. At most, maybe parts of the flight control systems are tuned with AI. But even then I'd be surprised if that's the case - computer controlled flight controls has existed since the 1970s.

>Of course in practice, we know that people's fears are not rational. Things like shark attacks and nuclear meltdowns scare people in a manner vastly disproportionate from their actual impact. It could be that the public has a lesser reaction to a plane going down with a person at the helm than an automated flight crashing.

In an analogy to self-driving cars...even if the math is sound, people will take issue until the technology is universally so far above a human driver that the accident rate is neglible.

I'm not sure about that. Even if the accident rate is half, that's still an immense gain - over 10,000 lives saved each year in the US. Even if it's equal (or slightly greater than equal), the productivity boost would be substantial. Do we really think that people with hour+ long commutes would reject a self driving car with half the accident rate of a human driver?

I don't know, probably not, but to your point about people's fear being irrational, I can see people strongly rejecting the idea of dying to a software glitch they don't understand.

Absolutely.

Most accidents occur when people are distracted, doing stupid things and at the same being sleep deprived and looking on the phone.

Imagine only producing accident rates twice as good as that, what a nightmare.

Human pilots are also a recipe for disaster. We put two of them in the cockpit for redundancy but they definitely aren't perfect. I'm not saying that we should adopt this right away but one day this will surely be the safer option. And a transition plan of having pilots on board for backup seems very reasonable.

I think I might trust an “AI” pilot more than a human one. Human pilots often fail to respond to unexpected conditions as well (see Air France Flight 447). Pilots have also gone nuts and crashed the plane intentionally (see Germanwings Flight 9525).

Something like this should at least allow airlines to use just one human pilot, who I doubt would do much 99% of the time. If there’s something that the computer can’t handle, the human could take over.

It's relatively easy to get the stats of fatal accidents that did happen, but how difficult is it to get the stats of accidents prevented by human pilots? Both are relevant to evaluating the impact of a 100% AI pilot.

I'm assuming the FAA in the United States has lots of records of incidents but I also suspect there could be many unreported incidents too. Companies can strong arm pilots into reporting internally or pilots may decide an incident was not worth reporting due to a potential impact on their career.

Does anyone know if there are regulations that require FAA reporting under specific conditions? What conditions qualify? How is this verified?

Dumb question but what is the benefit? I imagine pilots will still need to be there to take over during an emergency, so it won't save on pilot costs, right? And they can already lay back and relax during most of the flight so it won't really help with fatigue. Is the idea that autopilot can somehow take off with less risk than a human pilot? Can a human pilot take over sufficiently quickly if something goes wrong? Is pilot error during phases where such autopilot could've been running a source of accidents in practice?

I think the idea is you might be able to reduce the required number of pilots from 2 to 1 for some flights.

Or 0, eventually

Sure, but you get a huge amount of the labor cost reduction just by going from 2 to 1 and probably have a much easier time justifying it. Labor costs are already a pretty small part of air freight costs, for instance, so proportionally, the improvement in total costs (in both absolute and relative-to-total terms) from going to 1 from 2 pilots is about the same as going to 0 from 1.

EDIT: Just to illustrate... a 777 Freighter can fly a 5000 mile route at like 500 miles per hour carrying about 100tons of freight, I think in one shift so you only need a pilot and copilot. That’s 50,000 ton-miles per hour. Air freight revenue is about $1.40 per ton-mile, so in terms of revenue, you’re talking $70,000 per hour. Pilot and co-pilot wages are much lower per flight hour than you’d imagine, around $20-50/hour, so combined, they both contribute about $70 in direct labor costs per hour, so already just on thousandth, or 0.1%, of the revenue from the freight they’re carrying. Even eliminating the pilots doesn’t reduce costs much!

Much bigger difference for smaller and slower regional air freight routes.

Germanwings Flight 9525

...might have been prevented with self-flying...

had 2 pilots

Happened when pilot was alone

Cargo flights could fly without pilots, or use remote-control like a UAV

In addition to that-- is the pilot's salary a significant fraction of the cost of a flight?

Probably one benefit is having flights available and running 24/7, human pilots can only fly so many a day

The hard sell here will be convincing passengers it's safe enough. Not sure if I'd be willing to fly on a pilot-less flight even if the airline passed the cost savings onto me.

> convincing passengers it's safe enough

This is a pretty tough problem, different than self-driving cars because people see driving as something easy, but flying is seen as hard (just look at the difference in training and salary between a bus driver and a commercial pilot). Trained pilots/meat-computers will soon be worse than an automated system 99.99% of the time, but in those strange 0.01% corner cases the meat-computer really shines. There are several cases like Flight 214 (https://en.wikipedia.org/wiki/Asiana_Airlines_Flight_214) that computers would have helped, but there's still a few "Miracle on the Hudson" that any computer short of a general AI would have botched (https://en.wikipedia.org/wiki/US_Airways_Flight_1549). It will take a lot of work to get to where people will accept an automated flight without a safety pilot on board, and if you have to pay a safety pilot to ride along, you aren't saving as much money.

> even if the airline passed the cost savings onto me

HA! Like that will happen. The only time that airlines "pass on cost savings" to the customer is when a low cost carrier starts offering service for the same route.

There will probably be something like hundreds of flights in unpopulated areas, then overseas freight, then inland freight, then passengers.

$250,000 pilot cost per year (including their benefits and training, correct me if I'm too low) 261 working days 1 long-haul flight a day 300 seats in a 777 (disregarding class)

Comes out to ~$7 cost savings for two pilots per flight. Doesn't seem that impressive for a long-haul unless I'm missing something.

Just introduce it on a low cost airline, where the passengers literally don't care about a single other thing than price.

Just certified to fly without co-pilot would save money.

Maybe freighter drone at first. Might be easier to get certification to fly in drone mode between remote airfields at first.

A350 freighter seems to be under consideration. https://cargofacts.com/airbus-nears-launch-of-a350-freighter...

How many people will want to become pilots once flying (jets, fighters, etc.) is fully automated? Will it be difficult to find people to "take over" from the AI pilot in case of an emergency? I suppose you wouldn't need that many since it's unlikely there will be multiple emergencies concurrently.

I feel like this could backfire from a marketing perspective.

Imagine an airline integrating pilot-less flights, then a competitor starts using the slogan "We use real pilots, because we care more about your safety than increasing profits".

I wonder what this means in terms of security of the system? Planes are essentially missiles and allowing them to be accessible via a network could open the door to some pretty dangerous outcomes.

>"perform normally pilot-flown maneuvers entirely on its own"

It's the non-normal exception cases where pilot expertise is really critical

But can it serve Tomato juice?

Would you get on a plane with no pilot?

Hell no.

At least have qualified pilots who want discounted travel on board (being available for emergency take over pays for their flight—a new pilot takes over at their destination).

There won't be enough qualified pilots if most flights are autonomous.

So passengers have to demonstrate a successful landing in Microsoft Flight Simulator if they want to get a discount ;-)

Would you choose a plane with a pilot or without a pilot if a plane without a pilot is 5% cheaper? 20% cheaper? 50% cheaper?

Would you choose it if it were 0.5% cheaper?

Keep in mind that when you have a human pilot on board, they ultimately make all decisions about safety. And they have skin in the game. They are the ones who push back if they do not feel confident in weather conditions, or in the mechanical state of the aircraft to fly it. Would you prefer that those decisions be made by beancounters in a corporate office, who are more concerned about not having to refund people's tickets, than getting the plane safely from one destination to the other?

Compare the safety-obsession culture of commercial flight, versus, say surgeons - who, as a profession, aggressively push back on even trivial safety improvements - like checklists.

If the surgeon were killed every time they made a serious surgical mistake, you'd probably see a much better safety culture in that profession.

Would an airline choose a plane without a pilot if it’s 10% more expensive but 0.5% more likely to operate predictably on schedule?

What about the air-hostesses?

> The European manufacturer just completed flight testing for its Autonomous Taxi, Take-off, and Landing project in June after its flagship aircraft successfully navigated each phase of flight on its own as pilots simply watched.

Talk about training your replacement.

My little brother is starting pilots school soon, maybe I should talk him out of it.

You should because it's a terrible career, although really I think most pilots end up bailing out of the pilot career into good industries so in the end, no big deal. Let 'er rip and enjoy the view when you get it

Why is it considered a terrible career?

Low pay, horrific hours, limited opportunities to advance, doing the same thing a million times, dead-man’s shoes, having to work against a schedule, living out of corporate hotels, unionisation, union-imposed seniority rules, shrinking market, pending automation... what is there to like about it?

Top pilots with the best seniority who can play the union game talk about achieving the same pay an undergraduate intern gets in a tech company.

While there's doubtless something to what you say, I don't believe there are that many undergrad interns getting the pay of a senior pilot. You're thinking of an extremely select subset of tech companies.

Yes my comparison was with extremes... but I wouldn't bet against the average working pilot earning less than the average tech intern either.

Tech salaries aren't a reasonable baseline. It's an outlier.

Spend ~4/7 of your life in mediocre hotels far from home, work itself consists of sitting in a cramped box with even less to do than than the passengers on the flight, money can be good but not career-goals good... I can see it.

One reason it's not great is there are a lot of pilot wannabees that are stuck flying small commercial feeder flights, and they don't have much experience and training, and they get paid barely enough to live on. They hope to get on to one of the bigger companies (american, united, etc). The flights a few years ago that iced up in NY state (I think that was where) were sleeping on someone's couch for a few hours only before a flight, and they didn't have much training.

And what is your evidence that "most pilots end up bailing out of the pilot career"? And, what do you consider "good industries"?

I have some friends who love it, some who hate it and some who didn't finish and got a ton of debt.

If you decide to be a pilot reach out to real world pilots and ask for advice.

I have friends who did years being paid very little at local airports so they could "make it" to the airlines to get paid much better.

The biggest thing I observed from my 10+ friends is, being a pilot is completely measured by hours in the cockpit and certifications. At first you pay for them, then you get paid for them but not much because those people are paying for them, then you get paid a living wage for them because you're flying customers/cargo/etc.

My friends who are happiest as pilots either had the government/some other entity pay for their training, or were able to skip ahead in pay by getting directly to airlines.

> being a pilot is completely measured by hours in the cockpit and certifications

Yeap. So if you can fund the initial thousand hours (give or take) you should be ok. If you can't and you don't like small GA aircraft you'll have a miserable time trying to get paid to fly and have enough hours to be considered by airlines.

> My little brother is starting pilots school soon, maybe I should talk him out of it.

So he wants to be an airline pilot? Ultimately ferrying passengers? He should be ok. The industry is very safety conscious and moves extremely slowly. His children, if he decides to have some, may not be if they choose the same career.

I would be more concerned about smaller planes or freighters.

This is provided that this is something that he actually loves to do. Because otherwise not sleeping at home and weird hours will get old really quick. If he loves aviation you probably can't talk him out of it :)

He's doing the standard "change your mind every couple semesters" thing that most 20 year olds seem to do. Just a few months ago I was teaching him programming, now he's decided becoming a pilot is the thing to do.

It highly depends on what he wants to do in aviation and his motivations for working as a pilot. I can give my perspective as someone who had a full-time tech career and started the transition to working as a helicopter pilot several years ago. Some of this has already been stated in other comments - but thought I would add my opinions overall.

Signs you should not be a professional pilot:

* You're doing it for prestige. We're decades removed from the golden age of aviation. It may be cool to you to be a pilot - but most other folks don't view you as having a dreamy job. Don't be a pilot for how others view you, do it for you.

* You're doing it for the money. High paying(>$200k) jobs in airplanes are rare, and reserved almost only for captains at legacy carriers, and some corporate jet pilots. You need A LOT of experience and a lot of luck to get these jobs. And who knows how long these higher salaries will last. Some of the best paying($350k+) jobs in aviation are in cargo - UPS and Fedex - which are likely to be some of the first to automate away the pilot. Helicopter salaries are much much much worse. The highest pay I've ever seen was a heli-pilot with 35 years of experience and 30,000+ hours making $150k piloting a $22M aircraft. Given, he only works 6 months a year for that pay. But still a pretty terrible best-scenario pay rate.

* You're an adrenaline junkie who is looking for their next fix. If you want to get your license and go do crazy stunts out in the desert with no people around - be my guest. But you'll be eaten alive in most sectors of commercial aviation without a safety-first mindset. In some areas of aviation, sure, there's some risks. But they should always be calculated risks where the ends fully justify any added risk factor.

* You want to spend time at home. Most flying jobs are nomadic. It is hard on families, and isn't for those who are uncomfortable moving frequently and living the hotel life. Most pilots I know have been divorced several times if they've ever found someone who agree to be in a long term relationship with someone who moves so often and is around so little.

Signs you could consider being a professional pilot:

* You love flying, and nothing will keep you out of the cockpit. This is pretty much the only reason I think anyone should become a professional pilot right now. For many people there's something incredibly special about being in the air. As a hobby, aviation is incredibly expensive(doubly or triply so for helicopters than fixed-wing). So unless you have a career that can support your expensive aviation fix, you may actually be financially better-off flying for a living(as compared with a low/moderate salary job and flying away all your money). Finally, depending on what you want to do in aviation, there isn't always an equivalent option for hobbyist, and you have to go commercial to fly those missions.

Personally, I fall into the "I love flying" category. Even though my tech salary would have allowed me to fly small piston helicopters several times a week as a hobby, I wanted to challenge myself by flying more complex missions and aircraft than I could as a hobbyist - wildland firefighting and search+rescue specifically. This year is my first on a fire contract, and I can say without a doubt that for me the time/money/strain on relationships has absolutely been worth it. I love my job, and every day I'm off-site I wish I were working. However, there are some caveats. I'm single and enjoy being so. I seriously doubt I would stay in aviation if I found a long-term partner and/or started a family - the two do not seem compatible to me at all. Also, I still do contracting in tech for additional income. I know I have that as a fallback career if the aviation industry automates away the jobs I enjoy. So I'm not 100% sure if I would have done this without that fallback career plan.

With regards to automation in flying, I don't have a crystal ball. The FAA moves slowly. Very slowly. The only "fast" thing I've seen them do is add more regulations to UAV operators when they started becoming a potential danger to manned air-traffic. But I have never seen nor heard them being quick to loosen regulations or certify automation systems(in fact, FAA safety documentation discusses how automation can cause more harm than good when paired with a human operator - to highlight their thoughts on the matter).

But I would likely say that your brother will be able to find flying jobs for decades to come. Airlines will be slow to integrate fully automated cockpits due to regulation - and the first crash, I imagine, will likely roll back approvals for many years after(whenever it comes - we can't build perfect systems. And the NTSB would likely blame the system, even if pilots couldn't possibly have resolved the issues themselves). Also, if your brother is interested in any kind of back-country flying, I bet those will be even longer away from automation if they ever see it. Think flying people to rural airstrips in Alaska, flying float planes charters in the San Juans, or bringing supplies to villages in the middle of nowhere in Indonesia. The highly dynamic environments with unimproved strips and water landings would be a non-impossible but more difficult task than typical airports to automate. And the cheap labor available to these missions would likely not justify the additional expense of automation. Speaking of cheap labor - these back-country jobs pay absolutely garbage wages - but I personally know a couple former airline pilots who transitioned to rural flying missions because they were tired of flying a bus with autopilot and wanted to get back to what they considered "real flying".

Hope that mini-rant adds some valuable context for you!

Poor pilots sitting and watching inside the plane the moment they got obsolete. Cameras now, lidar and radar later for night and bad weather flights.

Do you mean the same systems driving Tesla cars to immobile objects on the road ?

Reply on news.ycombinator.com