LeSving wrote:
Why would an autonomous aircraft require ATC?
Because you want coordination. That won’t change. And ATC is moving to automation as well. One of the steps is shifting as much instructions on data links as those instructions will be easy to generate for an automated system and it will alleviate congestion on voice channels. Standardized instructions will also be machine readable so pilot only has to confirm it and the computer takes care of the rest.
alioth wrote:
I always thought it would be a spam filter that would first become self-aware due to the arms race between spammers and anti-spammers – which would of course be a pretty terrible existence.
Just imagine it would amuse itself by throwing out important messages and watching the havoc. 
2greens1red wrote:
I’d also be interested in hearing how a robot a/c would have managed the Sully incident.
That really depends if the programmers of the on-board software had considered ditching as an option to a forced landing if alternatives on land exist. If a ditching subroutine would have been implemented, then the outcome would have been identical. No, better: Without the human time constraints the computer would have worked the whole length of the ditching checklist and the aircraft would have stayed afloat for much longer. And somewhere I read that flight crews who tried the Sullenberger scenario in the simulator were able to glide into Teterboro. But only because they knew what was about to happen and began their diversion immediately. So theoretically, an autonomous plane might have made a successful landing at KTEB… who knows.
It is interesting to read the divergence of opinion between those who beleive it is around the corner and those that dont. I am a sceptic but some of the arguments are persuasive.
My good friend regularly takes CAT into Mexico City. He relates, it is always and interesting experience becasue the weather often features heavily with thunderstorms in the vicintiy. Inevitably there is the discussion between him (captain) and the co about whether to proceed. The co is sometimes reluctant, but with discussion is persuaded to have a look – and so they do, and sometimes a landing is possible, sometimes not. Human factors are heavily involved. As we know weather is not an exact science. I appreciate the computer will have the doubtless digitised input from the radar and control over the radar head and perhaps ground based radar and doubtless complex algorithims to make a deciison on the approach. It will be interesting to see if a computer is capable of providing a more precise and safe solution that the pilots. It is yet again another aspect that demonstrates how complex a computer controlled solution must be, and the amount of backup instrumentation that is required to ultimately replace the mark 1 eyeball that I guess still has its uses.
what_next wrote:
That really depends if the programmers of the on-board software had considered ditching as an option to a forced landing if alternatives on land exist.
This is the sticking point: a computer can only do what it was programmed to do, a human crewmember can be inventive and use judgement.
We can see the pitfalls in automation – e.g. AF447 – denied of sensor information, and with certain assumptions now being faulty, would a computer have done any better? (I doubt it).
The suggestion that you have a captain who normally does nothing but monitor also seems a bit dubious – if the automation were to become faulty, you now have a human who can use judgement but is horrifically un-current because they’ve not flown perhaps for months or even years. We can see the results of automation failures with a flight crew that doesn’t know how to actually fly with some recent crashes.
The thing is you’re going to probably need something that approximates a self-aware AI, then it’s going to start demanding rights and equal pay with the humans :-)
IMHO the technology required is so far ahead of the current state of the art that it is mind boggling to work out how anybody could make it work safely.
And not just in the cockpit. Within ATC too – they have been talking about digital messages from ATC loading autopilot preselects etc (but only upon a manually executed pilot confirmation, to prevent data link hacking) for many years.
Then you have the various airports to upgrade massively.
Who will pay for all this? Airlines of course!
alioth wrote:
We can see the pitfalls in automation – e.g. AF447 – denied of sensor information, and with certain assumptions now being faulty, would a computer have done any better? (I doubt it).
Another good example: An autonomous AF447 would not have flown through that thunderstorm in the first place (like everybody else didn’t either). Without flying through the storm, the sensors would not have failed. And if they had failed, an autonomous system would have taken all available sensor inputs into consideration, not just those from the pitot-static system, i.e. INS and GPS derived 3D velocities and accelerations and – most importantly! – the pitch attitude which those pilots obviously didn’t care about at all. Those two newbies were so fixated on the IAS reading alone that they completely lost the picture. AF447 would not have happened to an autonomous plane. With and without blocked sensors.
Peter wrote:
And not just in the cockpit. Within ATC too – they have been talking about digital messages from ATC loading autopilot preselects etc (but only upon a manually executed pilot confirmation, to prevent data link hacking) for many years.Then you have the various airports to upgrade massively.
A lot of that is within the capabilites of mode S and implemented almost everywhere in Europe already. And FMS programming via satellite link has been around for almost two decades and is already implemented in everything bigger than a B737 or A320.
what_next wrote:
And somewhere I read that flight crews who tried the Sullenberger scenario in the simulator were able to glide into Teterboro. But only because they knew what was about to happen and began their diversion immediately. So theoretically, an autonomous plane might have made a successful landing at KTEB… who knows.
According to the Wikipedia article:
The NTSB ran a series of tests using Airbus simulators in France, to see if Flight 1549 could have returned safely to their choice of LaGuardia Airport (LGA), either runway 13 or 22, or Teterboro Airport (TEB) runway 19. The test pilots were fully briefed on the series of events and maneuvers. The test pilots were only able to return successfully to either airport in eight of 15 attempts. The NTSB report noted that these test conditions were unrealistic: “The immediate turn made by the pilots during the simulations did not reflect or account for real-world considerations…” A single follow-up simulation was conducted where the pilot was delayed by 35 seconds: He crashed trying to return to LGA runway 22.
Airborne_Again wrote:
The test pilots were fully briefed on the series of events and maneuvers. The test pilots were only able to return successfully to either airport in eight of 15 attempts.
Which means that the fully autonomous plane would have been successful 15 times out of 15 attempts. Because it has zero reaction time. If knows it’s energy state every moment, the topography, the wind and everything else. It can continuously compute a plan B and a plan C in case anything should happen, and the very moment it senses a malfunction switch to one of those contingency plans. The Space Shuttle ascent was programmed like that 40 years ago. Apart from a few seconds after ignition of the solid rocket boosters it always had a safe abort option already worked out and ready for instantaneous activation. Either land back home, abort to a transatlantic alternate or abort to a different orbit. The only thing it could not cope with was the explosion of it’s external tank.
NB: One more thing: Sullenbergers maneuver only worked out because they had VMC conditions all the time. In IMC they would have had no chance at all. The autonomous plane doesn’t care about that.
what_next wrote:
Which means that the fully autonomous plane would have been successful 15 times out of 15 attempts.
That doesn’t follow at all. You assume both that the autonomous aircraft will at every moment be able to calculate the optimum action and that it has complete knowledge of the winds at every point in its projected flight path both in this and future instants. That is clearly not true. E.g. the (non) occurrence of a convective cloud on or near the flight path could completely upset the calculations and make all the difference in the world between making the runway or crashing some hundred metres (or more) short.
In this situation “almost making it” is no better that not making it at all. If you have the choice between attempting to make a runway which would succeed in 50% (say) of the cases and in the other 50% result in 100% fatalities in the aircraft and lots of fatalities on the ground on the one hand and on the other hand attempting a ditching in the Hudson which would have a high chance of success although some fatalities would be likely, then clearly the Hudson would be the better choice.
