[bicubic]

The problem with Tesla’s ‘autopilot’ is that it’s anything but.

Asking drivers to keep their eyes on the road and hands on the wheel while not steering guarantees that their attention will wander, because their brain isn’t getting enough stimulus to keep focused on the task.

I don’t know why it’s not clear to most people by now. The current Tesla ‘autopilot’ is simply more dangerous than manual driving because it harms human reaction time during emergencies.

Tesla is using legalese to blame people for this fully predictable effect when crashes do happen, but I suspect it’s only a matter of time before they’re forced to rebrand Autopilot as a lane assist technology which is all it is. Its only use as a safety system is to maintain control of the car is the driver becomes incapacitated, and safely bring it to a complete stop.

[cynix]

> Asking drivers to keep their eyes on the road and hands on the wheel while not steering guarantees that their attention will wander, because their brain isn’t getting enough stimulus to keep focused on the task.

From my experience, it's no different from your run-of-the-mill cruise control systems requiring, but not enforcing, that you keep your foot hovering above the brake pedal. I don't feel it's more dangerous than standard cruise control systems.

[cmsonger]

> Asking drivers to keep their eyes on the road and hands on the wheel while not steering guarantees that their attention will wander, because their brain isn’t getting enough stimulus to keep focused on the task.

Completely agree! When I'm driving her Tesla (no RSI issues here) I much prefer to use the adaptive cruise (which is really nice) and do my own steering for exactly the reason you cite. I have to pay attention anyway and steering keeps me from getting bored. Plus I do things like, you know, pass and keep myself from getting into bad situations that are many seconds ahead.

I only use AP to demonstrate it as a party trick to folks who have not seen it before. I half agree with the "use crowd sourcing to get AP training sets" but I completely disagree with how the thing is currently marketed -- even if (as you say) the legalese tries to cover Tesla.

Don't get me wrong: Her Model S is a spectacular car. I love it. But AP is somewhere between a party trick and a death trap unless, like my wife, you have serious problems holding the wheel for hours and are therefore happy to sit there and "supervise" what is a bad expressway autopilot.

[annerajb]

I am curious how do aircraft pilots got around this? Autopilot on aircraft work same as a Tesla drive a straight line where I aimed you have way less interaction than on the car. Yet pilots are able to take over autopilot and their responsibility during it's use is mostly looking out the window and occasionally switching frequency

[random4369]

There are two critical differences.

1: The handover latency (time from AP requesting handover to time pilot takes over) is measured in seconds to tens of seconds. AP is designed to give up a long time before any possible issues occur. Contrast this with cars on roads where the reaction times need to be in the sub-second range to avert crashes. If AP took a plane into terrain during poor visibility conditions and the pilots only got a second or two of terrain warning prior to a crash, such a crash would never be classified as pilot error on those grounds. Contrast this with self-driving cars where the autonomy frequently doesn't give up at all and the driver's awareness of the situation is the only thing to save them.

2: There are two operators on controls at all times. Recognising the limitations of human attention spans is one of but not the only reason for this being a requirement in civilian airlines.

Boeing has a whole design philosophy about making the operations of AP completely transparent to the pilots and failsafe. That means that all key controls (thrust, trim, stick, etc.) in the cockpit are physically manipulated by the AP so the pilots can see exactly what's going on. and more importantly that the controls represent the exact state of AP when the pilot takes over, so there are no unexpected sudden changes in input. The current generation of self-driving cars is a joke compared to the safety engineering that goes into AP systems.

[scott00]

One key difference is how much time operators have to figure out what's going on before action is necessary. If an aircraft autopilot fails, time before crash is likely measured in minutes, so the pilot taking 10 seconds to snap back into pilot mode still gives a pretty good likelihood of a favorable outcome. In a car, 10 seconds of no or bad driving on the part of the autopilot is pretty likely to cause a collision.

Even so, it's been implicated in crashes, notably Air France 447.

[etimberg]

I think the big differences are that:

* pilots have thousands of hours of training/practice

* there are 2 of them

* there is a lot less traffic in the sky / not making a turn every km

[bordercases]

> Asking drivers to keep their eyes on the road and hands on the wheel while not steering guarantees that their attention will wander, because their brain isn’t getting enough stimulus to keep focused on the task.

This is in HCI 101; we learned this by air traffic controllers crashing planes.