It is ironic because autopilots on airplanes are actually really simple, but somehow when the term is applied to cars people expect abilities that airplane autopilots have never historically had.
>In 1947 a US Air Force C-54 made a transatlantic flight, including takeoff and landing, completely under the control of an autopilot.
Airplane autopilots are already at the fully autonomous stage. They have the advantage of air traffic control removing obstacles from their path, but IMO this just means ATC is part of the autopilot system. An autopilot is something that gets you from point A to point B with no input from the pilot/driver, and if it's not capable of doing that reliably then it's not an autopilot.
Also, this is completely ignoring that a) autopilot doesn’t work at all without lots of infrastructure on the ground (try landing on autopilot when the airport ILS is down or absent, for example), and b) there’s a non-trivial set of flight conditions in which autopilot cannot be used at all.
People really should stop comparing plane autopilot to autonomous cars, the technologicalies share almost no similarities whatsoever. Plane autopilot doesn’t do any object recognition, AI, machine learning, adapting to the environment, etc. At the very best it will tell you when there’s another plane flying too close (based on an explicit location sent from the other plane or the ground), or when you almost run into the ground. That’s it. Everything else the autopilot does is just setting the flight controls to follow a preset path, and ensure the plane stays within the flight envelope.
I think this is what people expect from AP now, because that’s how it’s marketed: to do the heavy lifting 90+% of the time and when it fails and disengages It should give plenty of time to react.
Going forward AP will start to mean more and more “fully self driving”. Because that’s what some manufacturers claim it can do short of pesky legislation. So no supervision, no disengagement, just “piloting automatically”. And everyone pretty much has similar expectations from “automatically”.
Agreed, but it's Tesla doing that for marketing purposes that's the cause of these comparisons, not the people objecting to Tesla doing that for marketing purposes. Aircraft autopilot systems are relatively simple systems with a lot of instrument support and redundancy that relieve pilots of the need to hold a yoke in normal flight (with the latter bit being what the public tends to focus on). Tesla offers extremely complex software which makes more decisions based on less straightforward inputs and relies entirely on the driver holding the controls at all times in case it makes a bad decision.
That is not even remotely true. Pilots must babysit the autopilot and do. When flying on autopilot you still constantly check instruments to make sure that the plane is going where it should, how it should, as fast as it should. They sometimes have faults, and they have caused crashes. Pilot training is VERY clear that AP can never be relied on.
Eg: If you tell your AP to climb at 2000 ft/min, many will happily stall you trying to maintain that climb well above where your engine(s) provide(s) enough power to do so; if you tell your AP to descend to 3000 ft over sea level, most will happily fly you into a mountain that is in your way and rises to 4000 feet above sea level (in a better-equipped plane, TAWS will beep at you to tell you about this, but most APs will press on); if you tell your AP to maintain altitude, many will happily overspeed the aircraft in a large updraft, nosing down to avoid the climb.
And yet autopilot on cars is expected to intervene in those cases and far more. (Of course, cars are often in those situations far more often, which is partly why the expectations are different)
0: https://en.wikipedia.org/wiki/Traffic_collision_avoidance_sy...
You can get autopilots for boats as well and they can even do collision avoidance using radar, but its non obstructed 2D space with no real lanes in open water and lanes measured in 100's of feet in or more in channels.
Cars share such close proximity with obstructed curved paths that makes a safe "autopilot" orders of magnitude more complex which should be obvious. There is just no comparison due to the space constraints cars must navigate actively.