back to article NASA brainboxes work on algorithms for 'safe' self-flying aircraft

It's the fear of anyone who watches Snakes on a Plane and books a flight – what if your plane crashes? Now take a deep breath and imagine that you're travelling on a plane or rocketship with no pilot. A new NASA research project hopes to find ways to certify unmanned autonomous aircraft systems for safety. The easy part, says …

  1. regregular

    I wonder how this will work in terms of "acceptance of the public". Sitting in a semi-autonomous car is already very bewildering, but you have the pedals and steering wheel right in front of you and you know what to do.

    Now, with your butt planted in seat G02 and the plane taxiing to the runway, how many Xanax does it take to keep you calm?

    1. Steve the Cynic

      Where I live(1), the metro service has completely staff-free trains. No drivers, no on-train conductors ready to press a "close doors" button, nothing.

      Since 1983.

      (Yeah, yeah, I know, there's probably someone watching on CCTV, but you know what I'm saying. And the trains show evidence of having manual controls locked away under panels.)

      The trams (strictly speaking, metre gauge light rail because they are mostly segregated from the roads) have human drivers, but they have to cross roads laden with cars and pedestrians and without crossing signals of any kind.

      (1) Lille, France.

    2. enormous c word

      ...the future of autonomous vehicles can only be in aircraft.

      Autonomous cars have no real future - semi autonomy has a place for those unable to park competently or for enhanced safety. There will always be too many variables and too many *older* vehicles, cats, dogs, sheep, cyclists ....

      Semi Autonomous trucks or trailers with either a human or autonomous driver in the lead vehicle does make more sense, but the real future is in autonomous or semi-autonomous flight.

      While it is within the capability of most of us to master driving a car without any assistance, travel in a 3-dimensional space (flying) is something else entirely - it's only safe because there are really very few other vehicles competing for the same space - and lets face it mastering the controls and disciplines to enable safe piloting of an aircraft is beyond the capablity of most of us.

      The airways will only get busy when autonomous systems are doing the driving, and once they are busy, the only way to stay in the air safely will be with a good autonomous pilot.

  2. Gordon 10

    Waste of time

    This will happen when hell freezes over unless you are in the military - and even then remote piloting is more likely to be a thing first.

    Fleshbags prefer a fleshie to give them the illusion of control.

    Would rather see them work on safety interlocks for autonomous deployment of Sharks with Frickin Lasers on their heads. On the grounds its cooler and more likely to happen - Elon has is gonna run out of places to put his moola sooner or later.

    1. Anonymous Coward
      Anonymous Coward

      Re: Waste of time

      I personally wouldn't have a problem with a self flying plane. It isn't as though pilots never make mistakes.

      I think younger people who have been around computers all their lives will have little issue with this. It is the older crowd who will be resistant to self flying planes and self driving cars.

    2. Mark 85

      Re: Waste of time

      Actually, I think this will happen sooner than you think and for one reason... money. The airlines will push on this to cut the costs of pilots. Pilot's salaries and benefits are a pretty high cost for them. And no pilots and they one other less problem to deal with.. the pilot's union.

  3. JimmyPage Silver badge

    er ... "Failsafe" ?

    I have a vague memory that the concept of "Failsafe" is that a system can be put into a state which is intrinsically safe.

    For example, railway signals used to be DOWN for stop. In the event of the signalling system failing, all trains would just stop at the next signal. Leaving the system in a "safe" condition.

    I also believe there's no failsafe for an aircraft in motion. I.e: there is no way to set the controls to keep the aircraft permanently safe (even a car would just come to a stop when power is remove).

    1. Charles 9

      Re: er ... "Failsafe" ?

      For an aircraft, the failsafe goal is to LAND the plane. Now, given that, how you proceed to that goal depends on the conditions involved, but it should be noted that one universal condition about commercial airliners is that they need to be able to GLIDE in case all the engines fail.

    2. Francis Boyle Silver badge

      Your memory is somewhat misleading

      An individual component can (perhaps) be made failsafe but the system as a whole can only be made 'acceptably' safe. At the systems level you have something called 'safeworking'* - a set of procedures designed to prevent unsafe situations and to correct them as soon as possible if they do occur. That's what makes rail and air travel as safe as it is (which is very safe). And since "algorithm' is just a fancy word for "a set of procedures" it doesn't ultimately matter who or what is implementing them

      *That's the rail term at least.

      1. Charles 9

        Re: Your memory is somewhat misleading

        I don't know the exact word in the airliner industry, but there are damn good reasons there are lots of talks about "checklists" and "procedures" in regards to airliner operations. The idea is to take as much as you can into consideration so that when something does happen, there's a procedure already in place for it.

  4. Flakk

    I'm a Little Confused

    Aren't planes already largely self-flying? Aside from handling taking off, landing, and adjusting trim during the flight, doesn't the autopilot already fly the plane?

    If comparing the concept to self-driving cars, I think there are probably far fewer environmental concerns to deal with (airways are probably quite orderly when compared to highways). The mechanical variables are a completely different story, as an aircraft is vastly more complex than an automobile. And of course, the impact of a failed automated system in an aircraft would be far greater.

    Drone pilots can fly their aircraft from half a world away. Could airlines realistically assume remote control of an airliner in a similar fashion, should the automated systems fail?

    1. Anonymous Coward
      Anonymous Coward

      Re: I'm a Little Confused

      What do you mean by "automated systems fail"? Just like there is redundancy built in for flight controls, there can be redundancy built in for the autopilot. If you mean a software failure - a "I don't know how to handle this" kind of situation, it would have to alert the ground at least a minute in advance of reaching such a situation. Even with a pilot able to assume remote control standing by, he'd need some time to determine the current state of the aircraft before deciding what to do. Latency would be another issue - even if the ground pilot wasn't halfway around the world, 50 ms latency might be enough to screw up any attempt at a Sully-like heroic save.

      I don't think it would be practical, even if you had pilots standing by at the controls, monitoring every takeoff/landing (on the theory that almost all "oops I don't know what to do now" situations will occur then, rather than during ordinary flight at 35kft.

    2. Steve the Cynic

      Re: I'm a Little Confused

      "Aside from handling taking off, landing, and adjusting trim during the flight, doesn't the autopilot already fly the plane?"

      I refer you to the Unreliable Source:

      https://en.wikipedia.org/wiki/Hawker_Siddeley_Trident#Avionics

      First automatic landing: 1965. First "blind" landing: 1966

      OK, agreed, not strictly the autopilot in the classic sense, but automatic, anyway.

  5. John Smith 19 Gold badge
    Unhappy

    The problem is what to do if things go wrong...

    "Autoland" has been available since the late 60's and was certified with a failure rate of 1 in every 10^9 hours of operation. Possible with hard wired systems and triple redundancy and necessary because the number of "tests" mounts up fast with the number of commercial aircraft (of a certain type) flying and taking off day in, day out.

    But what happens if the sensors the aircraft is using to sense its surroundings fail? What happens if one of the engines fails? What happens if the controls fail but you can still control the aircraft through engine throttling (BTW following NASA tests that is now a flight computer mode on some A/C)? What happens if you can't get the undercarriage down?

    Most of these are survivable throughout most of the flight (even an engine fail on takeoff within limits) but the system has to diagnose the fault and decide on a course of action..

    Safe solution. Code detection tests and actions for all known (or at least probable) failures and trigger a re-evaluation if sensor data changes (but remember the sensor data may be the fault and the system may be operating normally).

    Riskier solution. System that applies logic to the problem IOW a form of AI.

    BTW NASA does this already for some of its probes and scientific satellites The trouble is can such a system a)Diagnose what's happened b)Develop a plan to mitigate it c)Implement it fast enough to save lives?

    Fixed altitude/heading/speed cruise you could do with an 8 bit micro controller.

    Or you just do what they do with microlights and fit a whole vehicle parachute and just fire that.

    1. Christoph

      Re: The problem is what to do if things go wrong...

      And what happens when you hit a flight of geese during take-off - can the auto-pilot make the decision to try to land on the Hudson River? (Though it may be able to do the actual landing once the decision is made).

      1. Charles 9

        Re: The problem is what to do if things go wrong...

        The trick there is that the plane's computer would need to be aware of its location, surroundings, and situation. With the current level of technology available, it is possible to provide all three. If it knew it had just taken off from New York City and was evaluating possible emergency landing sites, picking the Hudson for a water ditch could be reached under the right instructions (a return was impossible, and anywhere else on land was risky due to the dense population. A water ditch near a shore would be the preferred course in such a situation--not too close to people yet not too far away, either). Even for things not explicitly seen before, we should be able to tell the computer to use heuristics or other techniques to find a best fit (which if you think about it is how humans respond to novel situations).

      2. John Smith 19 Gold badge
        Unhappy

        "And what happens when you hit a flight of geese during take-off"

        "can the auto-pilot make the decision to try to land on the Hudson River? "

        Well those are the questions, aren't they?

        And of course can it do so in time to make a difference?

      3. Anonymous Coward
        Anonymous Coward

        The Sully save

        And if a plane can't do the Sully save, so what? What percentage of actual airline pilots could have pulled that off? I don't know the number, but it is surely well under half. Could Sully even do it again if the situation presented itself a second time? Even with all the skill his years of experience gives him, there may have still been some luck involved - i.e. what if there was a tour boat in the river right where he was going to touch down but he wasn't able to see it when he started his glide path because it was under a bridge?

        There's no way you can make a computer able to recover 100% of incidents, but human pilots can't either. Like humans, computer pilots will get better over time as they handle more and more situations, but the experience comes from all computer pilots back to them all so soon it will be the equivalent of a Sully with 2000 years of experience, and no human pilot will be able to touch it. Yes, there can and will be software bugs, but humans do stupid things too - get drunk, fall asleep, hit the wrong button and so forth. Maybe have three separately developed software pilots, and they have to agree or 'majority rules' to minimize the impact of bugs.

      4. JeffyPoooh
        Pint

        Re: The problem is what to do if things go wrong...

        Christoph highlighted, "...make the decision to try to land on the Hudson River?"

        1) Decision to *immediately* switch on the APU (outside of any preconceived Checklist)

        2) Decision to assign the co-pilot the role of going through the Emergency Checklists

        3) Decision to turn out over the river, getting away from the city core

        4) Decision that both return and Teterboro were too late / too far

        5) And yes, finally the decision (conclusion?) that, "We'll be in the Hudson."

        The Airbus flight control system, to its credit, did help him to ditch very precisely.

  6. AnoniMouse

    Multi-vehicle collision avoidance

    Algorithms for reliably avoiding collisions between multiple, autonomous (i.e. not centrally controlled) moving vehicles (or, in 3D aircraft) are not plentiful, especially if circumstances where scalablity in numbers of vehicles (beyond two or three) vehicles is necessary.

    All the challenges of complexity, with horrendously challenging scalability.

    1. Anonymous Coward
      Anonymous Coward

      Re: Multi-vehicle collision avoidance

      Who says they won't be centrally controlled? They'll have "air traffic control" with the big picture view telling them where to go, and the individual autopilots will execute it. Just like human air traffic control tells pilots where to go. Avoiding collisions should be way easier than on roads which are essentially 1D or 1.5D instead of 3D.

      All those near misses (and some not-misses) in the air are because a human screwed up. The bar isn't that high for software to beat them.

  7. Your alien overlord - fear me

    Suddenly hitting wind sheer will need faster processing than seeing a brickwall and trying to avoid it. All of a sudden the plane is in zero gravity as are the passengers (and it is not a nice feeling) and seconds later all the instruments are saying you're 3000 feet lower than you were before. So will the plane realise or just assume multi-instrument failure?

    1. Nolveys
      Thumb Up

      @ Your alien overlord - fear me

      All of a sudden the plane is in zero gravity as are the passengers (and it is not a nice feeling)

      I've had that happen a few times and it's always been the highlight of the flight.

  8. NoneSuch Silver badge
    Terminator

    It will work perfectly.

    Until it doesn't.

    1. Anonymous Coward
      Anonymous Coward

      Re: It will work perfectly.

      But then, humans aren't perfect, either...

  9. John Jennings

    Its not self flying you need worry about

    its the self crashing....

    Hacked plane == missile.

    most airports are near or in cities

  10. JeffyPoooh
    Pint

    Acceptable short-term Use Case follows in 3, 2, 1...

    Commercial Aircraft presently have TWO (2) pilots. With a self-flying aircraft, the two pilots could easily be reduced to one pilot. This transition can happen over a much shorter timeframe than the more ambitious target of zero meat-based pilots. It achieves 50% of the savings with (maybe) 5% (<- WAG alert) of the 'Acceptable Level of Safety' requirements.

    (The next prediction after that is as follows: As Satcom becomes increasingly available, fast, cheap and reliable, then the one backup pilot might be sitting in a shed somewhere, on call. Covering a thousand aircraft. [Polar flights will need IridiumNEXT Satcom, due to polar coverage.])

    1. Anonymous Coward
      Anonymous Coward

      Re: Acceptable short-term Use Case follows in 3, 2, 1...

      I don't think satcom can be used in an airliner situation: too much lag. And before you mention remote-piloted drones, most are short-range and controlled from the point of takeoff. The longer-range ones IINM have more limited roles because even a split-second lag can be problematic in a twitch situation.

  11. aviator

    Not anytime soon...

    As a longtime pilot and current airline pilot, I can't begin to tell you how many times the computers on my aircraft have glitched. I've had complete system failures when ALL data was lost, both flight and navigation, the autopilot and throttles disconnected and the plane had to be flown manually till the non- flying pilot could reload all data. One pilot could not have accomplished the task. I have flown aircraft with satellite data link since it was introduced years ago and data links fail frequently, (including three times on my flight home from Tokyo this week), especially at high latitudes. Though obviously not the same, anyone who uses a computer or cell phone knows how this works and how frequently things need to be rebooted or signal is lost. Two humans are there because, like it or not, things ALWAYS go wrong and humans, with our amazing brains, are best equipped to deal with the unanticipated.

    1. Anonymous Coward
      Anonymous Coward

      Re: Not anytime soon...

      gotta love the reg, it's nice to have someone who obviously knows what they are talking about!

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like