Boeing Studies Planes Without Pilots, Plans Experiments Next Year (seattletimes.com)

"Boeing has begun researching the possibility of commercial-passenger jets that will fly without pilots, using artificial intelligence guiding automated controls to make decisions in flight," reports Seattle Times. The company is planning experimental flights, without passengers, for next year. From the report: "The basic building blocks of the technology are clearly available," said Mike Sinnett, former chief systems engineer on the 787 Dreamliner and now vice president at Boeing responsible for innovative future technologies, at a briefing before the Paris Air Show. "There's going to be a transition from the requirement to have a skilled aviator operate the airplane to having a system that operates the vehicle autonomously, if we can do that with the same level of safety," Sinnett said. Sinnett said Boeing's research is driven by the pilot shortage worldwide that is only going to become more acute. In the next two decades, Boeing forecasts a demand for about 40,000 new commercial jets, roughly doubling the world fleet.

Comments

no pilots no passengers

[turkeydance]

no warning.

meh

[zlives]

maybe automated-trains should be a proven tech first.

Re:meh

[MightyYar]

Behold! The future!

Re:meh

[aristotle-dude]

maybe automated-trains should be a proven tech first.

Vancouver has been living in the future since 1986. The Skytrain system is the longest automated train system in the world. https://en.wikipedia.org/wiki/...

Re:meh

[lucm]

Airbus planes are basically flying without pilots. There's people in the cockpit but they can't override the computer in case of emergency (since French engineers know everything) and can only do minimal things within specific guidelines enforced by the computer; they're just slightly more in charge of the planes than flight attendants.

Re:meh

[walterhpdx]

This. Very much this. Airbus aircraft generally scare the shit out of me.

Re: meh

[Anonymous Coward]

I am Airbus A320 pilot and also a software programmer, albeit not a good one. In any case, you are wrong. Airbus does let you override the computer. In fact, the computer is quite dumb. It only has protections such as bank angle limits. It's still easy to mess up. Even the ECAM (electronic diagnostics and problem resolutions), is still dumb. For example, it can lead you to disconnect two IDGs if blindly followed. Would you want to be over the ocean without electricity?

I love the A320/321 but don't place too much faith on how smart these systems are. They only automate the most benign of tasks.

The Airbus A320 is safer than a 737 but it still requires quite a bit, if not more, knowledge.

Re: meh

[MightyYar]

The Airbus A320 is safer than a 737

Isn't this only true if you include the older 737 models from last century, from before the A320 was available? I think they are both remarkably safe planes, probably without enough statistical data to differentiate their safety records.

Re: meh

[torkus]

Clearly that system is poorly designed and implemented. Funny how everything in aviation needs to be tested to the 3rd degree yet some obviously broken things still pass muster.

I don't think (or hope!) that anyone here is suggesting taking a brain dead 'automation/decision tree computer' and letting it fly a plane without a pilot.

But a properly designed system? That's different. Heck, you could probably use some of the machine learning from IBM Watson and let it watch a few million hours of pilots flying and it could extrapolate almost any conceivable flight situation with the best response. Use that response set to program your new flight computer :)

IANAP (pilot) but I've done a small bit of flying in the cockpit of small planes. As I understand it, flying is technically complex, but generally quite rigid and structured (i.e. straight-forward with a given body of knowledge). Driving, as best as I can interpolate, has FAR more variables and many of those are much less predictable. I'm curious on your take regarding the comparison.

Cargo

[JBMcB]

I would think this would be a near no-brainer for cargo flights. Probably less so for passenger flights.

I would consider flying a robo-flight if they installed an authentic HAL 9000 eye on the cockpit door, if for no other reason just to see it.

Re:Cargo

[bugs2squash]

no, it would be one of those rectangular watson lava lamps, I wonder what they're called.

Re:Cargo

[rtb61]

So they claim to have invented a secure wireless computer network system that could never ever be mass hacked, remember, greedy, cheap, lowest tender and hack one, hack them all. So automated planes the perfect false flag system for endless wars, they choose who they get to blame the attack on based on current corporate profit goals ie resources that need stealing, munitions that need expending and replacing, any hint of nationalising of essential services (the perfect corporate blackmail tool).

Re:Cargo

[HornWumpus]

Cockpit will be gone. Seats with front views will be very expensive though. Even though the 'windshield' will be tiny windows.

Computer checks pilot

[aberglas]

If you have a pilot at all, they need to actually fly the plane, or they will deskill. No half-smart AutoThrotle that can cause crashes like the Air Asia one.

And then we need a smart AI system to monitor the pilot and warn them if they are doing something stupid. Like trying to land the plane miles short of the runway.

If the pilot does not respond the autopilot can disconnect the controls.

It used to be said that you need a pilot and a dog. The pilot to feed the dog, and the dog to bite the pilot if they touch the controls. But the Autopilot can the job of the dog as well. Maybe electric wires in the seat.

[AutoThrotle -- when flying a small plane, one constantly monitors air speed on descent. But large planes have autothrotles that are like cruise control and do this for the pilot. But if they autothrotle is set to the wrong mode, then nothing is monitoring the air speeed. which has led to several crashes.]

Re:Computer checks pilot

[dknj]

which has led to several crashes

citation needed

Re:Computer checks pilot

[R3d M3rcury]

"deskill"? Is that like roadkill, but with your desk?

(Yes, I know, It's de-skill. As in lose skills. But I read it "desk-kill" and thought it was funny. Like the pilot will suddenly go postal sitting behind a desk or something.)

Re:Computer checks pilot

[fustakrakich]

citation found!

Re:Computer checks pilot

[Anonymous Coward]

Nothing like the AI disconnecting the controls from a pilot who is still trying to fly a plane. But the AI decided it knows better.

Ever tried to indent in Word when it has decided you don't want to do that?

Re:Computer checks pilot

[Jeremi]

citation found!

You call that a citation? This is a citation.

Re:Computer checks pilot

[Anonymous Coward]

And then there is asiana at san fransisco. And Turkisch at Amsterdam. And a Thompson incident in England (no fatalities that time).

Re:Computer checks pilot

[4im]

Nothing like the AI disconnecting the controls from a pilot who is still trying to fly a plane. But the AI decided it knows better.

Ever tried to indent in Word when it has decided you don't want to do that?

No need for AI for something like this to happen, it already happened. See The untold story of QF72.

I think the very idea of an autonomous, civilian plane is ridiculous at this time. I only need to imagine one of those freighter 747s flying over here each and every day, manipulated into crashing into town. The damage could go much further than even 9/11. The only context in which the autonomous plane might make sense is in a military context - and I do believe drones are already close enough to this.

Re:Computer checks pilot

[afxgrin]

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

If they just let go they probably would've been fine.

Re:Computer checks pilot

[religionofpeas]

Yes, it's a good thing that planes with human pilots cannot be manipulated to fly into buildings or mountains.

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

Re:Computer checks pilot

[Anonymous Coward]

Trust the computer. The computer is your friend.

Re:Computer checks pilot

[torkus]

In this and the other citation provided, the crash happened because the pilots fought the autopilot (and made plenty of other mistakes). Had they NOT tried to compensate for a situation they didn't fully understand, they might have had a rough time until the auto-systems got them on a level flightpath but they'd be alive.

The difficulty here isn't skill loss due to automated systems, it's pilots and autopilots not being able to interpret what the other is doing. If the pilots in this example had simply let the plane go it would have (presumably) just not landed. The would have had to explain the aborted landing but ... no dead people DESPITE incorrectly setting and auto-system.

In the other, it's almost exactly the same. A child playing in the cockpit during flight (idiot pilot/father there) did what's necessary to disengage some of the auto-flight systems. The resulting situation was greatly exacerbated by the pilot not knowing what was happening and him fighting with the auto-systems caused the crash.

You can't have two masters in a system. Despite these autopilots NOT being designed to overcome gross stupidity, they WOULD have handled both accidental caused situations if they had been allowed to (based on the reports and conclusions therein). Similarly, had the pilots been prepared (and able) to completely remove all flight automation they probably could have recovered the planes as well.

Plus autopilots don't get bad ideas of inviting children to cockpits or flipping the wrong switch in a moment of tired/stupid/drink/bored/sick/unconscious.

Doomed from the get go

[i_ate_god]

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

Now imagine there were no humans who could pilot the thing on board. When that eventually happens, and it will happen, people will become far too afraid to fly in AI only flights.

Re:Doomed from the get go

[MightyYar]

I can counter that with a suicidal pilot - something that machines don't yet aspire to. I'm betting that if we sat here all night posting examples of human vs. machine errors in aviation, you'd run out of material first unless we go waaaay back.

Re:Doomed from the get go

[ShanghaiBill]

When that eventually happens, and it will happen, people will become far too afraid to fly in AI only flights.

Pilot error is one of the most common reasons for crashes. AI auto-pilots don't need to be perfect, they just need to be better than humans.

So start them out in cargo planes, continue to improve the tech, and once they pass humans in reliability, no more human pilots.

Re:Doomed from the get go

[Obfuscant]

Pilot error is one of the most common reasons for crashes.

Pilot error is one of the most common results from NTSB investigations, but that doesn't mean pilot error was actually the cause. PE is what the NTSB hangs things on unless they can find good evidence of something else -- because the manufacturers of any hardware or software blamed for a crash have lawyers who make good money defending them, while the dead pilot cannot pay anyone to defend him.

Re:Doomed from the get go

[0123456]

"Pilot error" is what they call it when the situation is so bad that the autopilot doesn't know what to do, hands control back to the pilot, and the pilot crashes.

Re:Doomed from the get go

[dgatwood]

I can counter that with a suicidal pilot [wikipedia.org] - something that machines don't yet aspire to.

Give them enough time around airline passengers and they will.... :-D

Re:Doomed from the get go

[Anonymous Coward]

So now instead of a suicidal pilot you get disgruntled technicians/engineers/hackers who can auger in a plane full of people while they sit back safely on the ground eating some popcorn and watching the show. Personally it would be a LONG time before I'd get on a jet that didn't have a flesh and blood person at the helm who was going to die right along side of all of the other passengers if they failed. There is a lot to be said for the human tendency of self preservation. I think this kind of technology would be great for cargo, as long as it has hard coded limiters which would prevent it from flying over or landing/crashing in populated areas.

Re:Doomed from the get go

[MightyYar]

disgruntled technicians/engineers/hackers

The problem with your argument is that those people could sabotage a plane right now, so simply adding more automation does nothing to help or hurt that scenario.

Hudson River

[xbytor]

Let me know when these AIs can land a plane on the Hudson River after a massive bird strike.

Re:Hudson River

[Ichijo]

Or a missing wing!

Re:Hudson River

[Anonymous Coward]

The Hudson river crash was due to pilot error.

They hit the birds quite high, 2,000' by memory. They took a full 30 seconds to decide to turn the bird around, by which time it was too late. Fortunately, the river happened to be nearby. (Wikipedia is your friend.)

Any computer would have got that plane back to the airport. But then now news story.

As to an actual landing on a river, not that hard. You need some moderately clever vision software to identify rough vs smooth crash sites. Tricky but doable. Or just use cameras to let a remote pilot handle just the emergency.

You didn't read the wiki page. It says -

The incident came to be known as the "Miracle on the Hudson",[3][4][5] and a National Transportation Safety Board official described it as "the most successful ditching in aviation history."[6] The Board also rejected the notion that the ditching could have been avoided had the crew immediately returned to LaGuardia or diverted to nearby Teterboro airport.

Re:Hudson River

[aaarrrgggh]

How does computer land plane back at airport without power for said computer? You can say "batteries" but that is a failure mode you also need to address.

Also, the "smart" part about landing was to do it near where the ferries are stationed so you have a chance at rescue. There isn't really a lot of empirical data on forces as you land on water...

As graceful as the landing looks in the videos, I am sure a computer would have been able to manage air speed and angle of attack better, it is the unforeseen secondary issues that really made the difference though.

Re:Hudson River

[ShanghaiBill]

Let me know when these AIs can land a plane on the Hudson River after a massive bird strike.

Most human pilots couldn't have done it. An AI pilot, if it was trained with water landings as a criteria, could.

Whenever people try to point out the weakness of AI, they always pick weird corner cases. But the thing is, these are often where an AI excels. An AI can be trained on thousands, or even millions of simulated water landing scenarios, and replay them over and over until it handles them properly. A human pilot will likely have zero experience.

Once the AI pilot is trained, then the "water landing module" can simply be copied into every plane with a USB thumb drive. Human pilots need to be trained one by one.

Re:Hudson River

[fluffernutter]

If AI is so easily trainable on any kind of edge case, why do they have such simple problems such as missing red lights or going down one way streets the wrong way? It would seem reasonable that they have that down before we should believe every edge case is covered.

Re:Hudson River

[aberglas]

And I might add that the root cause was that the pilots did not see the large flock even though the weather was good. A very simple computer based vision system would have picked them up in plenty of time to avoid. As would a good set of pilot eyes that were actually looking out the window.

(No, the planes do not fly so fast that vision is useless. They are flying at about 100m/s, and it is easy to see a LARGE flock of geese at 1000m, just look up in the sky sometime.

Re:Hudson River

[aberglas]

I think that ending up near the ferries was mainly luck. There is no evidence that he manoeuvred to be near them. Just went straight in. Fortunately, the river is long and in the direction he was then flying, so the plane just went down in a straight line and eventually met the water.

Re:Hudson River

[HornWumpus]

Where he landed was dictated by the planes energy and avoiding bridges.

Re:Hudson River

[Anonymous Coward]

So, like, who are you? Bob Hoover?

Re:Hudson River

[postbigbang]

Or a busy airport pushes planes out with a low ceiling on climb. Some nameless fool is using a RaspberryPi3 and doing pirate radio that nulls out the vision system. Engines fail, maybe both. The plane is overloaded with fuel to begin with, for a long flight. Suddenly the options are thin.

Siri? What do we do now? Several tugs and ferries are on the water landing zone, and that unpleasant silence of no thrust is pounding in whose ears? No ears.

It will take a very, very long time before AI can replace human pilots, fallible as they are. What happens when the ILS goes down in the middle of a windy thunderstorm? I've had wind push the tail so hard that we were landing sideways, but lived to tell about it. Feel free to search on windy landings, especially one made of a day at DUS to decide just how much you trust a Boeing program on the rudder pedals.

So many industries are pushing to get rid of transportation drivers because of their supposed costs that it's almost a mantra among the MBAs in transportation companies, who have nickled and dimed us to death. Those pesky pilot unions, always wanting more..... yet many pilots get paid less than bus drivers. It's all about playing to the greed of airlines, who loathe the next political disaster that craters their stock.... and their pension funding (looking at you, United).

Re:Hudson River

[aaarrrgggh]

That isn't how Sully tells the story; they had two or three potential locations to land and it was a deliberate choice according to him.

Re: Hudson River

[Anonymous Coward]

The computer would have misdiagnosed.

Re: Hudson River

[Anonymous Coward]

100m/s eh? I dont think so

Re:Hudson River

[Anonymous Coward]

Intriguing — can you tell us the story of your last engine failure on takeoff? Or do you only specialize in Monday morning quarterbacking of other peoples' experiences?

Re:Hudson River

[dgatwood]

The three most likely reasons are:

• Overreliance on map data (that turned out to be wrong) instead of visual signals.
• Cutting corners in training to get to market sooner.
• A human overriding the computer.

Re: Hudson River

[Nkwe]

100m/s eh? I dont think so

Why not? 100 meters per second is about 225 miles per hour. Take off speeds are slower and cruising speeds are faster, so shortly after takeoff, 100 meters per second doesn't sound unreasonable.

Re:Hudson River

[Anonymous Coward]

Funny you should pick "missing wing" as a criteria, since exactly this scenario has been demonstrated by Rockwell Collins:
https://youtu.be/xN9f9ycWkOY

Re:Hudson River

[sad_]

The Hudson river crash was due to pilot error.

They hit the birds quite high, 2,000' by memory. They took a full 30 seconds to decide to turn the bird around, by which time it was too late. Fortunately, the river happened to be nearby. (Wikipedia is your friend.)

i've only seen the clint eastwood film, there the delay is explained as followed:
- there is a checklist they have to go through when something happens that covers all options
- it takes time for humans to decide what to do

that is not human/pilot error, that is just human behaviour. it takes time to react. in fact the pilot skimmed huge part of the checklist because it would take too much time.

Re:Hudson River

[Anonymous Coward]

Speed is life. Altitude is life insurance

Every Part 121 airline has crew resource management procedures which dictate you do nothing until you figure out what the problem is. Overly quick, incorrect action has killed more than delayed, appropriate action.

Cactus 1549 was in an awkward phase of flight where you have enough life, but you don't yet have all the life insurance you need. Also, don't forget that a double engine flame out as the result of a bird strike had never happened. You can't windmill start a turbo fan without 300 kias. Trading altitude for airspeed works when you have the altitude.

Re:Hudson River

[Anonymous Coward]

The NTSB used flight simulators to test the possibility that the flight could have returned safely to LaGuardia or diverted to Teterboro; only seven of the thirteen simulated returns to La Guardia succeeded, and only one of the two to Teterboro.[85] Furthermore, the NTSB report called these simulations unrealistic: "The immediate turn made by the pilots during the simulations did not reflect or account for real-world considerations." A further simulation, conducted with the pilot delayed by 35 seconds, crashed.[86] In testimony before the NTSB, Sullenberger maintained that there had been no time to bring the plane to any airport, and that attempting to do so would likely have killed those onboard and more on the ground.[87]

Re:Hudson River

[drinkypoo]

I've had wind push the tail so hard that we were landing sideways, but lived to tell about it. Feel free to search on windy landings, especially one made of a day at DUS to decide just how much you trust a Boeing program on the rudder pedals.

But that's just the thing, isn't it? Any modern car is better at threshold braking than a human, and moreover can do something that no human with less than three legs (real ones, mind you) can manage: brake each wheel independently while also managing the throttle. (you can build a brake pedal with two master cylinders behind it such that the heel and toe each affect a different end of the car, a trick used in rally racing.) A plane is going to be much the same thing, only moreso; it can manage each control surface independently, and if you put an antilock brake at each wheel, can also exert yaw control through both the control surfaces and the wheels. And MEMS accelerometers can react faster than the inner ear, and even mundane archaic ABS can detect wheel slip faster than a human can, so even if you had enough limbs to manipulate all the controls at once, and enough brain power to manipulate all the limbs at once, you still wouldn't be able to react as fast as a computer.

So many industries are pushing to get rid of transportation drivers because of their supposed costs that it's almost a mantra among the MBAs in transportation companies, who have nickled and dimed us to death.

And yet, the cause of the majority of incidents is pilot error.

Re:Hudson River

[PPalmgren]

The reason they added 30 seconds was because they went through their checks to attempt to get the engines back online before diverting. The simulations were diverting almost immediately after the bird strikes because they knew the bird strike resulted in failure, when most bird strikes do not and most bird strikes allow engine restarts.

Re:Hudson River

[postbigbang]

And they have to do all of this in near or actual real-time, summoning the actions of possibilities of thousands and thousands of actions, and do it right, just like a pilot with several years experience, and make the sensation essentially a normal day for passengers.

I never said pilots were infallible. I'd rather die in a crash caused by a pilot, than AI.

You can pitch ABS as an example of modern design wisdom; and as a subsystem, it's true-- usually. It adds lots of complexity and most drivers don't know how to use it.... and technicians can't repair them well.

Multiple control surfaces in play are a candidate for systems control because of multiple concurrent variables. It's merely a capex vs opex shift in where you put the investment-- sophisticated control plane systems (which are also proprietary vendor-to-vendor, and not heterogeneous in a meaningful way) or in pilot training and opex salaries. I'd rather employ pilots. I like humans; I am one.

Re:Hudson River

[drinkypoo]

You can pitch ABS as an example of modern design wisdom; and as a subsystem, it's true-- usually. It adds lots of complexity and most drivers don't know how to use it.... and technicians can't repair them well.

This is patent nonsense. First of all, most of the time you don't bother to repair an ABS module. You replace it. If it's new, you replace it under warranty. If it's old, you can get a used part. If you go to a shop, of course, they're not dicking around with used parts. They're just going to buy a reman and slap it in. They follow the instructions and everything works fine.

The only time this is not the case is when you need a magical scan tool that your local indie shop doesn't have in order to bleed the brakes. My A8 doesn't have ESP so you just bleed it in the usual fashion, but for Audis that do you need a $250 cable. For Mercedes (including Sprinter vans) you need an outrageously expensive scan tool (low to mid thousands) or if you are very lucky you can sometimes pick up a $500 device which can do the job. And yeah, that's stupid, but it's not the technology's fault. It's the automaker's fault.

I'd rather employ pilots. I like humans; I am one.

I'd like to use whatever does the job best, enhancing passenger safety. At the point at which they can show that the computer is better at doing the job, I'd rather have the computer.

Re:Hudson River

[postbigbang]

I snicker. I restore cars. There are tools I have and can't afford/don't need. Dealers don't understand ABS well, and their techs don't know how to repair it well. I have lots of admittedly anecdotal evidence to point to. Some of the 1990s ABS designs are worse than none at all. Today, it's a bit better. And some vendors train their techs better or put better instructions into Mitchell.

Let me branch to driverless cars, another problem looking for a spot marked X. In 48yrs of driving, I've seen too many accidents, and many of them would have been compounded by driverless vehicles smashing into the results of them, or perhaps not stopping to aid injured people. Here again, I have no comfort with the possibilities of driverless AI.

Yes, I like passenger safety, too. We have zero data on pilotless accidents because we don't keep track of drones, even in the military, in a way that produces feedback to the vendor/pilot as regards safety components, strategies, etc. A failure in ILS can be devastating if no active pilot can take over.

The ILS glide path is a thing of beauty. Well-understood, and highly protected-- because if it fails, we're back to manual landing. Will AI know when to abort a landing? Do we make AI do 1000 touch-and-goes before we certify it? Make it land at coastal airports where wind shift and shear are common? What's the regimen? How can we know the success of AI? We've been using human pilots for 115yrs... and AI? How long of a base and foundation does it really have?

Re:Hudson River

[drinkypoo]

I've seen too many accidents, and many of them would have been compounded by driverless vehicles smashing into the results of them,

The driverless vehicles are going to notice the incident sooner, and they're going to be less willing to take risks than the average human driver. They're also going to be communicating with on another via V2V. V2V will probably become mandatory for all cars, providing even more information about traffic accidents. I'm not thrilled about that either, but I do expect it.

or perhaps not stopping to aid injured people.

They certainly will not stop to aid injured people. They will be in constant communication with a control center which will be staffed with humans who will get an exception notice that vehicles are reporting something they're not expecting along the highway, and then they will determine the nature of the event from video footage collected by the vehicles and alert the authorities. Their primary job is not to alert the authorities, but it's cheap to do so. Their primary job is to provide useful routing information for the vehicle network.

The ILS glide path is a thing of beauty. Well-understood, and highly protected-- because if it fails, we're back to manual landing. Will AI know when to abort a landing? Do we [...etc.]

Those are all valid questions, the whole paragraph in fact, which is why I didn't bother quoting more of it; I'm not going to argue with it. I'd start with the same sort of certification as a human pilot as a baseline, and I literally do not see any reason not to expect an order of magnitude more repeat testing before I took it seriously. I want to give it every chance to exhibit bizarre behavior. But humans also sometimes exhibit bizarre behavior, so it doesn't have to be perfect. It only has to be better than a human. Even twice as good would satisfy me.

Re:Hudson River

[Neuronwelder]

Oh the AI have this long of a base: $$$$$$$$$$$$$$$. Money talks. (and kills)

Re:Hudson River

[Anonymous Coward]

The Hudson river crash was due to pilot error.

They hit the birds quite high, 2,000' by memory. They took a full 30 seconds to decide to turn the bird around, by which time it was too late. Fortunately, the river happened to be nearby. (Wikipedia is your friend.)

i've only seen the clint eastwood film, there the delay is explained as followed:
- there is a checklist they have to go through when something happens that covers all options
- it takes time for humans to decide what to do

that is not human/pilot error, that is just human behaviour. it takes time to react. in fact the pilot skimmed huge part of the checklist because it would take too much time.

It's also a case ripe for automation.

The checklist can be easily translated into a an algorithm and a computer will process it faster than a human.
Most checklists are O(c) algorithms juts being a nice bit list of if/else clauses, while complicated ones might be O(n) if they actually have some loops where a solution is attempted then the result re-assesed.

If a human can get though a checklist in 30 seconds by skimming it (possibly missing something important), I'd expect an algorithm to do it in under 1 without taking any shortcuts.

In this case simulation indicate that might have been enough to get the plane down safely on a runway, and failing that as 30 was adequate for a ditch, 1 would be even safer as you get a whole 29 extra seconds to optimize the approach.

Re:Hudson River

[Anonymous Coward]

V2V communication directly between the AIs controlling the cars? What could possibly go wrong? Just wait until the first exploit is found...

Re:Hudson River

[Anonymous Coward]

Here is a thing that's important for autonomous cars:

Visual signals must ALWAYS win if there is a discrepancy between what the map says and what the camera says. And that means, you can just get rid of the detailed maps and just use the normal ones everyone uses for their GPS in the car.

The problem is to make the car AI really understand what it's seeing. Just picking out all the details is not enough, it needs to understand the whole context.

Re:Hudson River

[Anonymous Coward]

"During 2004, pilot error was pointed to be the primary reason of 78.6% of disastrous GA (general aviation) accidents, and as the major cause of 75.5% of general aviation accidents in the US": https://en.wikipedia.org/wiki/Pilot_error

I don't understand why people think that humans are so superior, when in fact they are pretty often worse than random guess.

Re:Hudson River

[aberglas]

And that was a very stupid thing to do.

Near the ground, their first thought should have been where they can glide to and land. By 2,000' they have passed the dead zone and should easily make it back. Now, if there had been the river in front of them, and the airport behind, and only 1,500', then they could have been well excused for going for the river. But the river actually led back to the air port, so there was no conflict.

Engines do not stop without good reason, so it is very unwise to bet on being able to restart them. And they did the right thing there, with the co-pilot trying to get the engines going while the pilot flew the plane. But the pilot stuffed up monumentally in not turning promptly. 30 seconds is a long, long time. Count it out and see.

I suspect that very few of the people commenting on this thread have ever flown any airplane. Otherwise this would all be obvious. And I do not know why the investigation let the pilot off. But a turn even after 15 seconds would have got them back OK. I would have thought 5 seconds would be heaps after a dual engine failure to figure out what to do. And remember, pilots are supposed to be alert during take off and landing, precisely because things can happen relatively quickly.

Which again raises the question of why they did not see the geese.

Re:Hudson River

[dgatwood]

Visual signals must ALWAYS win if there is a discrepancy between what the map says and what the camera says. And that means, you can just get rid of the detailed maps and just use the normal ones everyone uses for their GPS in the car.

Not necessarily. A sign could get hit by a car and bent in such a way that it appears to affect the wrong road. But yes, in general, you're correct.

Is It Really The Future?

[HannethCom]

Automated trains have been around since 1967. Though I didn't know ours is the longest ALRT in the world. The first part of Skytrain was criteria 1986.

I would project that there will be a pilot

[Anonymous Coward]

But we won't need the copilot on most flights, and the pilot will only need to be in the seat during takeoff and landing, while wearing a radio earpiece the rest of the time. Should free up pilots for the other flights.

That's fine

[PopeRatzo]

Surely we don't need pilots.

And don't call me Shirley.

Re:That's fine

[fustakrakich]

I want Ted Striker and Elaine Dickinson as my pilots

Edge cases are hard

[Hentes]

It's always easy to automate most of a problem, but edge cases tend to be really hard to solve. Yes, the autopilot can fly the plane 99.9% of the time, but the pilots are there for the 0.1% when it can't.

Re:Edge cases are hard

[Kjella]

It's always easy to automate most of a problem, but edge cases tend to be really hard to solve. Yes, the autopilot can fly the plane 99.9% of the time, but the pilots are there for the 0.1% when it can't.

Well, from what I understand they hand over full control to the pilots given sufficient failure because they're there and supposed to be experts, but in many cases it could have continued and in many cases pulled through. Or the damage is so extensive the pilots can't control the plane or don't understand the situation themselves. Or the pilots don't know what to do in these error conditions and don't know how to fly either. For example Air France 447.

According to the final report, the accident resulted from the following succession of major events:

• temporary inconsistency between the measured
• speeds, likely as a result of the obstruction of the pitot tubes by ice crystals, causing autopilot disconnection and reconfiguration to alternate law;
• the crew made inappropriate control inputs that destabilized the flight path;
• the crew failed to follow appropriate procedure for loss of displayed airspeed information;
• the crew were late in identifying and correcting the deviation from the flight path;
• the crew lacked understanding of the approach to stall;
• the crew failed to recognize that the aircraft had stalled and consequently did not make inputs that would have made it possible to recover from the stall.

If they'd just let the computer carry on from a best guess air speed based on thrust, altitude and angle of attack, they'd have been fine:

At 02:10:34, after displaying incorrectly for half a minute, the left-side instruments recorded a sharp rise in airspeed to 223 knots (413 km/h; 257 mph), as did the Integrated Standby Instrument System (ISIS) 33 seconds later(the right-side instruments are not recorded by the recorder). The icing event had lasted for just over a minute.

Every time shit like this happens, we improve the safety systems. And so every time the pilots become less and less accustomed to actually handling anything other than normal mode. Which is when we don't need them.

Re: Edge cases are hard

[Anonymous Coward]

Sensors provided erroneous data to the computers giving the pilots false data to react to. The computer was programmed to disengage since it too cannot fly based on faulty and contradictory data.

Yes... better pilots would have recovered but computers would have done worse had they followed guidance from sensors.

Re: Edge cases are hard

[dgatwood]

That depends on how the computer is programmed. One could presumably change the failure mode so that it continues without changing anything until it gets valid data, which would have probably prevented the disaster. Or one could make the failure mode be an emergency descent, which would at least have avoided the stall, in all likelihood, and maybe avoided the crash, too. And a computer could ostensibly use other data like GPS to determine its ground speed and altitude, and use that data as a crude estimation of air speed in the absence of proper data, which is something a person isn't likely to think to do in such a crisis, much less be able to do the computation properly.

In other words, it's hard to say what the computer could or could not have done. A computer might have done much worse, but if they anticipated the situation when constructing the training data, it would probably have done much better. Either way, once the model is trained on such a scenario, it would almost certainly do better.

Re: Edge cases are hard

[Anonymous Coward]

In other words, the computer did none of the sensible things you suggest. Instead it was programmed to give nonsensical high-level alerts to the pilots, causing them to panic and finally crash.

The problem, to me, is that the computer /interpreted/ the incorrect data as a stall. It should have given the pilots a medium-level alert "inconsistent speed data - possibly dangerous - please check immediately".

But no, the computer had come to a conclusion - the wrong one, and the pilots didn't have the time to understand what was going on.

How many more wrong programmed conclusions are we going to find?

Business Oportunity

[SPopulisQR]

Clearly this is an opportunity to install front row seats with panoramic view...

Re:Business Oportunity

[HornWumpus]

Windshields are heavy and can break. They will install litle windows oblique to the front and still charge a huge premium.

Re:Business Oportunity

[fustakrakich]

More likely there will be no windows. Just not worth the expense.

Re:Business Oportunity

[budgenator]

We sat on the tarmac for 8 hours waiting for the windshield to be replaced and cure, it kind of sucked, but not as bad as have a windshield shatter at 550 MPH and 35,000 feet.

Re:Business Oportunity

[HornWumpus]

If people, as a group, could be packed together like sardines without any windows the airlines would have done it 50 years ago.

Perhaps if they sedate the claustrophobic passengers.

Next - Planes without passengers

[Anonymous Coward]

Oh my!

Electriccal Fires et. al.

[Geodesy99]

See http://www.skybrary.aero/index... "I ... can feel myself ... going ... Dave"

The pilot of an aircraft has many legal, emergency, and crew leadership duties which go beyond the actual piloting of the aircraft.

Being a pilot has been described as long periods of boredom punctuated by seconds of sheer terror.

The pilot shortage is a red herring, like any other occupation, if you pay people commensurate to their educational investment, skills, knowledge, experience, and continue their training. The airlines have had a pretty good ride up until now because they piggy-backed on the military as a pipeline.

Re: Electriccal Fires et. al.

[Anonymous Coward]

The airlines have had a pretty good ride up until now because they piggy-backed on the military as a pipeline.

So Still confused ..... There is a pilot shortage because This pipeline from the military shutdown? How?

Re: Electriccal Fires et. al.

[Anonymous Coward]

No there is a shortage because they do not pay them enough. In the past they paid more than the air force so people moved from the military to private companies . Now they do not.

Re: Electriccal Fires et. al.

[Anonymous Coward]

Also there is no shortage in the EU because airlines pay more. Thanks to unions.

May be closer than you think

[Sycraft-fu]

While I'm not ready to go all in on AI controlled planes yet (or let's call them something else like Expert Systems, they aren't real AIs) I think starting to test is very valid. We are able to design systems with very good decision making capabilities these days. It is conceivable that we will soon be able to make them on par with humans, even for extreme cases like 1549.

It is certainly an area worth putting R&D in to.

HAL 9000 murdered all its passengers

[drnb]

You realize the HAL 9000 murdered all its crew and passengers? ;-)

Re:HAL 9000 murdered all its passengers

[zamboni1138]

Spoiler alert!

What windscreens or windows?

[drnb]

What windscreens or windows? Everyone will get a display that can be configured to show forward, side, rear, etc cameras. Windscreens/windows add complexity and cost.

Re:What windscreens or windows?

[HornWumpus]

People get too claustrophobic. Not all of them. If what you say was psychologically possible, they would have done it years ago.

Other than flying...?

[pubwvj]

I wonder how AIs react to hijacker demands?

Re:Other than flying...?

[DigiShaman]

They don't. But then again, I suppose that's the entire point, now isn't it?

Re:Other than flying...?

[Nkwe]

I wonder how AIs react to hijacker demands?

That AI software has already been written. In a previous life it was called "Clippy". "It looks like you are trying to hijack this plane. Would you like to..."

Re:Other than flying...?

[pubwvj]

Yes, that was my point. This could make hijacking a thing of the past.

Re: Other than flying...?

[Anonymous Coward]

There ate already only a few hijacking attempts a year. Was there one this year?

However, you could hack the plane and automatically run it into something.

Likely follow fly-by-wire adoption

[drnb]

In the minds of many fly-by-wire was also removing the pilots, having the computer fly the aircraft. However after decades of use in military aviation the general public accepted it for passenger aviation. The complete removal of pilots will likely follow a similar adoption, it will need to have a highly successful decades long history of use in military aviation first.

Plus, even with removal of human pilots from the aircraft there may be the capability to remotely pilot the aircraft.

Re:Likely follow fly-by-wire adoption

[torkus]

The military has been using remote piloted and autonomous drones for a while now.

Also consider that most large, modern plans include
Autopilot
Autoland
Autobrake

I don't know of any auto-takeoff but as I understand it that's a simple process (full thrust, maintain heading, when speed > preset, rotate and ascend) with the typical emergency-abort criteria.

Mainly what pilots seem to do is make announcements, talk to towers, and request a flight path change if there's turbulence. None of this is especially complex to automate. Oh, and if you DO automate the planes then you open up the option of automating some (though not all) of the tower operations...which are known to be the most stressful, demanding, and difficult jobs in aviation.

Now...i'm still stupefied as to why the MTA (NYC Subway etc) and other cities trains still run things completely manually.

Re:Likely follow fly-by-wire adoption

[drnb]

The military has been using remote piloted and autonomous drones for a while now.

I was referring to top tier fighters and bombers. Current drones and RPV seem to mostly be of a lesser tier of aircraft.

I don't know of any auto-takeoff ...

F/A-18 carrier launches.

i'm still stupefied as to why the MTA (NYC Subway etc) and other cities trains still run things completely manually.

Government employee unions with a sympathetic city government.

Heinlein hit this nail on the head

[Anonymous Coward]

I will never fly on a plane if the pilot isn't also on-board with me. He may not be able to as good a job as the computer and may cost more than a ground-based drone pilot, but in an emergency I know he'll do his damndest to try to save both our lives.

Re: Heinlein hit this nail on the head

[Anonymous Coward]

What makes you think your life is worth adding $5 on the ticket for a pilot? Besides, it would mean we have to keep the cockpit and can't have First Class Prime View seating or anything else we want to monetize that space. If you don't like it, go by train or go by ship.

Re: Heinlein hit this nail on the head

[RabidReindeer]

That "whoosh" over your head was not an Airbus!

My prediction

[Vasheron]

It won't fly.
*chuckles*

Simple

[Anonymous Coward]

No Unions!

Jajajajajajajajajajajajajaja

Checklists and prepared-for emergencies.

[Anonymous Coward]

The thing is... a computer can "work a checklist" way faster than two cooperating pilots can. Good.

So we get rid of the "we need time to work the checklists". The plane just radios to the airport: "I have an engine problem, MAYDAY I want to land back on runway 05, in 13 minutes, 23 seconds."

But in un-prepared emergencies, some pilots have taken the right decisions for a safe landing. For example that plane in hawaii that blew its top. It landed way overspeed because of control problems when slowing down.

So in the "normal" cases you get a bit better, but in the exceptional cases, things get a bit worse.

Re:Checklists and prepared-for emergencies.

[Archtech]

So in the "normal" cases you get a bit better, but in the exceptional cases, things get a bit worse.

That remark reminded me vividly of Frank Herbert's comment (in "The Dragon in the Sea"/"Under Pressure") that "there is no such thing as a small accident on a submarine". I suspect that "a bit worse" is a huge understatement (except in the sense that each of us can only die once).

How to settle the argument.

[Anonymous Coward]

Let autonomous cargo planes fly for a couple of years, and watch the performance data. Once they exceed the average human pilot in hazardous situations (crosswind landings, engine out, unexpected traffic, etc.), making the switch will be a no-brainer.

Front seat

[BESTouff]

On a plane without pilot I would pay extra to have a front seat and enjoy the view during the whole flight.

Progress!

[Archtech]

Oh good. In the past there have been incidents when the computers apparently took over an aircraft and locked out the pilots.

http://www.smh.com.au/good-wee...

Now there won't be any pilots to be locked out, so the aircraft can just destroy itself in its own preferred way.

There is no Pilot Shortage

[Anonymous Coward]

World wide airlines have no problem employing pilots. However, in the US companies pay them like busdrivers with worse working hours. As companies mainly compete over price, there are only a few options to stay afloat. The easiest is to reduce salaries. This work especially well when there is no market wide union negotiating salaries where every airline is bound to pay.

When the computer has full control and hurts pax

[Anonymous Coward]

We really want computers to fully control planes?

Read this story:

http://www.smh.com.au/good-weekend/the-untold-story-of-qf72-what-happens-when-psycho-automation-leaves-pilots-powerless-20170510-gw26ae.html

WIkipedia:
https://en.wikipedia.org/wiki/Qantas_Flight_72

I would not get on a plane without a pilot and co-pilot.

sounds familiar

[Anonymous Coward]

so... drones?

no way

[e70838]

The main role of a pilot is to handle failure of some parts of the aircraft. There is a huge number of stories where the pilots have been incredibly creative and have found ways to save desperate situations. IMHO, if the pilot does not risk his life, he may be less inspired.

An oldie but a goodie

[mccrew]

For years now, they have been saying that the cockpit of the future will consist of a pilot and a dog. The pilot is there to feed the dog, and the dog is there to bite the pilot if he tries to touch the controls.

Not over my house!

[laie_techie]

I'm glad there's no passengers on the plane, but that still risks everyone on the ground under the flight path.

AI Pilot?

[Neuronwelder]

Yea, I can see this ending well.. Oh, by the way, do home owners have accidental jet crash insurance yet? If so, how much more, and more, will they pay for it?

"overhead" fee special offer

[cosmicl]

save $10! $50 gets you 1 carryon bag and 1 pilot for your flight ($60 if purchased separately)

So no one is going to bring up...

[GESUS]

9/11 on this subject?
When it is finally relevant?

Now, I am not an alarmist or back the "war on terror" ie "license to do what ever". But this is actually a solution that does not kill people.

Regarding implementation.
First of, this system would first replace the two pilots rule.
After millions of hours of training and testing it would move to cargo flights.
THEN to passenger flights.

It must be self contained.