Live data from Hacker News

Study: Tesla Autopilot misleading, overestimated more than similar technology

mercurynews.com

151–159 of 159 posts

Re: Study: Tesla Autopilot misleading, overestimated more than similar technology

#151
post #5

Earlier quoted context omitted.

I see the use of “autopilot” name bring brought up a lot. Thinking about what an autopilot in an airplane does, labeling Tesla’s system autopilot is actually pretty apt. It’s unfortunate that there’s a disconnect between what autopilot means colloquially vs what it actually does on a plane.

> what an autopilot in an airplane does ...Flies the plane safely with zero input from the pilot (hands off the yoke) indefinitely? Lands the plane on its own [1]? I'm well aware that autopilot systems need input directions for desired altitude and air speed, but the semantic distinction people try to draw between "real" autopilot and Tesla's autopilot capabilities have never made sense to me. [1] https://en.wikipedi…

> the semantic distinction people try to draw between "real" autopilot and Tesla's autopilot capabilities have never made sense to me.

Typical airliner autopilot flies along a list of 3D waypoints and manages airspeed to avoid falling from the sky or overstressing the airframe. It's a fairly trivial system since it does not have to find its way around a complex, constantly changing environment like a car does. It will happily fly into mountains, other airplanes or dangerous environments like thunderstorms.

All decision-making depends on pilots. Autopilot itself is nothing more than a simple cruise control that relieves pilots from things like manually maintaining constant altitude over six hours on a transatlantic flight. Any hobbist with a Raspberry PI can built a similar GPS-based waypoint following and speed scheduling for a car, but it's useless for road traffic, because roads are not straight, long empty stretches that can be navigated by driving from one waypoint directly to another waypoint 100 miles away.

Car autopilot must be able to make decisions (follow the road, react to obstacles) to be useful, and that makes it fundamentally more complex than any existing airliner autopilot.

Re: Study: Tesla Autopilot misleading, overestimated more than similar technology

#152
While it's true that with current roads the self driving car software solution may not converge to an acceptable solution, it's also true that as the software gets better municipalities will start to modify their streets and roads to assist the software. At that point the combined solution will very likely converge to a "better than human" result.

Re: Study: Tesla Autopilot misleading, overestimated more than similar technology

#153
post #146

Earlier quoted context omitted.

It’s “Ok” until you’re involved in a fatal accident and Tesla blames you for not having your hands on the wheel. Doesn’t matter that Elon himself has demoed it hands off, each of the autopilot fatalities was the fault of the driver.

Did you know there are 1.5 million auto-related deaths per year around the world. And in 7 years around 23 associated with Teslas? And none of the autopilot accidents affecting other people?

How many times does it have to be repeated, those stats compare apples and oranges. Luxury, modern, new vehicles driven by statistically more wealthy drivers have a much lower accident rate than most cars. When corrected for these factors Tesla’s safety rating is abysmal.

None of the autopilot accidents affected other people? I guess if by affected you mean fatalities, but 2 of them smashed into tractor trailers, so that definitely affected others. There are multiple reports of Teslas on autopilot running into stationary rescue vehicles as well, thank goodness they haven’t been fatal.

Re: Study: Tesla Autopilot misleading, overestimated more than similar technology

#154
post #16

Earlier quoted context omitted.

They are thinking the same thing as people who drive drunk. People can be idiots. You have to actively disable safety measures in order to get the car to drive for any kind of extended period without applying force to the wheel. Anyone caught doing this intentionally should be cited for impaired driving same as a DUI. Then there are the people who legitimately fall asleep at the wheel and AutoPilot saves their lives,…

Yeah, the first time I drove one, I was holding the wheel at the bottom, with two hands, but not tight enough. The car gave me 3 warnings, then disabled autopilot for the rest of the journey . I had to get off the freeway, stop, get out and lock the car, then it would let me use autopilot again...

Your mistake was that it detects rotational force, not your grip.

Rotating the wheel just a nudge with one finger is enough to let it know you’re there. You can also rotate the volume or cruise speed adjutant wheels —- sometimes I will tweak it up and down a click (meaning you end up with no change in setting) instead of rotating the wheel.

Re: Study: Tesla Autopilot misleading, overestimated more than similar technology

#155

Earlier quoted context omitted.

Babysitting Autopilot that can unpredictably steer me into a concrete barrier would be more demanding than driving a car that is more predictable.

driving a car that is more predictable which you drove into the baricade, or got rear ended by or side swiped someone on the exit . the avg driver is more deadly than any tesla on autopilot . By the stats .

Which stats? The only statistic I'm aware of Telsa claiming is that Tesla Autopilot is safer than all other road vehicles combined.

Re: Study: Tesla Autopilot misleading, overestimated more than similar technology

#156

Earlier quoted context omitted.

driving a car that is more predictable which you drove into the baricade, or got rear ended by or side swiped someone on the exit . the avg driver is more deadly than any tesla on autopilot . By the stats .

Which stats? The only statistic I'm aware of Telsa claiming is that Tesla Autopilot is safer than all other road vehicles combined .

Tesla’s are really expensive to repair which impacts insurance costs. But, some companies have self-driving car discount‘s such as: https://www.joinroot.com/blog/tesladiscount/

Suggesting a net benefit exists relative to the base model without self driving capabilities.

Re: Study: Tesla Autopilot misleading, overestimated more than similar technology

#157
post #148

Earlier quoted context omitted.

Not always. The computer focuses better, but isn’t yet as capable. When things get dicey and the computer abruptly hands control back to a zoned-out human, they won’t have the same situational awareness as if they had been driving all along. OTOH, driving also tires humans out, which leads to errors and accidents too. Figuring out the right balance is non-trivial.

Maybe. It may be the case as other commenters in this thread have said that the automation actually lets them prepare for unexpected events more than normal because it handles the mundane steering and speed adjustments.

That seems very, very unlikely to me.

I can imagine that being able to relax on the highway makes subsequent high-stress city driving a little more bearable, but people are really bad at going from zoned-out to action mode quickly.

Re: Study: Tesla Autopilot misleading, overestimated more than similar technology

#158
post #148

Earlier quoted context omitted.

Maybe. It may be the case as other commenters in this thread have said that the automation actually lets them prepare for unexpected events more than normal because it handles the mundane steering and speed adjustments.

That seems very, very unlikely to me. I can imagine that being able to relax on the highway makes subsequent high-stress city driving a little more bearable, but people are really bad at going from zoned-out to action mode quickly.

I'm challenging the assumption that driving aids cause drivers to "zone out" any more than normal driving. Driving automation has been common for decades in the form of cruise control. I don't see anyone making the claim that using it makes you "zone out". Auto Pilot is just an evolution of this that handles the steering too. Drivers still need to drive and be engaged.

Re: Study: Tesla Autopilot misleading, overestimated more than similar technology

#159
post #146

Earlier quoted context omitted.

Did you know there are 1.5 million auto-related deaths per year around the world. And in 7 years around 23 associated with Teslas? And none of the autopilot accidents affecting other people?

How many times does it have to be repeated, those stats compare apples and oranges. Luxury, modern, new vehicles driven by statistically more wealthy drivers have a much lower accident rate than most cars. When corrected for these factors Tesla’s safety rating is abysmal. None of the autopilot accidents affected other people? I guess if by affected you mean fatalities, but 2 of them smashed into tractor trailers, so…

"Luxury, modern, new vehicles driven by statistically more wealthy drivers".

As far as risk aversion, I would say this group and Tesla owners are near opposites. Add in the completely different torque and breaking behavior. A completely different cockpit experience.

Post reply on HN