Hacker News Reader

Comment by avalys | Hacker News Reader

Autopilots in airplanes have nothing to do with autonomous driving.

Airplane autopilots are so simple they they can be implemented with entirely analog circuits. They are simple feedback control loops that operate off a single scalar variable in each axis.

There is ample time for the pilot to correct any “mistakes” the autopilot makes.

There is no expectation that the autopilot operates independently of the pilot’s supervision and judgement.

In terms of decision-making and autonomy and liability - an airplane autopilot is really analogous to cruise control in a car, nothing more. It operates the flight controls to maintain simple parameters chosen by the pilot, and that’s all.

View on HN

Maybe the analogy would be more accurate comparing to autoland systems? https://en.wikipedia.org/wiki/Autoland

AFAIK automatic landing systems just get the plane into the right 'landing parameters range', but won't take action if anything unforeseen happens (like a blocked runway).

Right the Garmin Autonomi Autoland system is intended only for emergency use to get the airplane on the ground if the (single) pilot is incapacitated. It's not intended for routine autonomous flight.

I think the earlier comments are referring to CAT 3 auto land systems using technologies like ILS, which are used routinely and are good for 0ft visibility. That is, approach in cloud down to the ground

Not only that, Pilot training assumes Autopilot can disengage at any time.

On top of that, aviation autopilots operate in one of the most controlled and stable environments. Ground based individual traffic is the polar opposite of that.

As aviation people eloquently put it: https://en.wikipedia.org/wiki/Big_sky_theory I think there should be a correlative 'small road theory', if only to underscore the difference.

It seems the non-existence of a small-road-theory is in itself an problem when it comes to self-driving cars...

Which is why (imo) Autopilot was a spot-on name for Tesla's initial driverless feature.

Except the general public seem to massively over-estimate how capable aircraft Autopilots are, and then expect Tesla's Autopilot to live up to their inflated expectations.

the general public doesn't fly planes, so to them every flight is getting on a plane, falling asleep, and magically teleporting from one city to another. They have no perspective on just how much an autopilot doesn't do.

I'm talking solely about liability when the system that you sell as being able to operate on behalf of a human gets it wrong.

Comparisons of the systems is irrelevant to the very simple point I made - putting your money where your mouth is garners trust.

As for your claim about autopilot being simple, and humans having ample time to correct its mistakes, well, two 737 MAXs and one Air France flight are very straightforward counterarguments.

Yeah, that's not autopilot. FLCS, fly-by-wire, drive-by-wire are all different things. Autopilot is really specific. If you want to discuss this productively, you should probably learn the difference.

I apologise that I got the precise terminology wrong when I used a term that most humans understand to mean "when the computer controls stuff your plane is doing on your behalf".

My biggest regret about getting it wrong is that it gives pedants something to fixate on and split hairs about, in lieu of engaging with the actual point.

Ease up a bit - not everyone has to be an aviation expert.

That's true, but some aircraft are now equipped with TCAS and GCAS which will automatically seize partial control from the pilot and maneuver to avoid some types of crashes. So far I think those systems have been 100% reliable but the manufacturers would be liable if they caused a crash.

> Airplane autopilots are so simple they they can be implemented with entirely analog circuits.

For reference: https://www.398th.org/Research/B17_AFCE.html

Reply on news.ycombinator.com