Featured Tesla Q2 2024 Safety report:

So probably people are using autopilot more on highways where you get more miles without crashes?

 

Like others adaptive cruise control and lane keeping, Autopilot can work at slower speeds, but without the stop reading abilities of FSD, most people likely only use it on highways and stretches of road without stops.

 

This is called "selection bias". It has been pointed out for years, but we still aren't getting the data transparency needed to gauge or address it.

 

Would drivers who use AutoPilot while it is under investigation by NHTSA be subject to liability lawsuits in the event of an accident?
This was some time ago but before this system was under investigation:

Tesla driver is charged in a deadly crash involving Autopilot : NPR

Tesla Autopilot Lawsuits (klinespecter.com)

Tesla must face vehicle owners' lawsuit over self-driving claims | Reuters

 

Autopilot do read stop signs and red lights but will only warn you, not stop. I wish they included that functionality to manual driving too. It could prevent so many t-boning.

 

I use it on highways and regional roads unless the traffic speed on those is faster than 10 km/h over the speed limit (hard limit by AP on roads other than highways), then it's just TACC. I tried the Volvo lane assist on those same regional roads and after a few km, I disengaged it as it can't follow the curves on those roads.

 

I think the selection bias has been with each single autopilot crash, people say autopilot has been killing people. Instead with and without autopilot are much safer than the average car. We have a whole thread assuming that bad assumption.

Most teslas are newer cars, and older cars may be much less safe. Tesla buyers appear to be more affluent which may make them safer drivers. What can not be said is autopilot is unsafe because it has been involved in a tiny percentage of accidents where the driver should have taken over. I don't know if teslas are more or less safe than a new toyota with safety sense, but certainly autopilot is not killing the huge percentage of people in the database.

Investigation has nothing to do with if it was subject to lawsuits. Normally in wrongful traffic deaths people go after the driver and car company, as the car company has more money. But the incidents I have read about the driver should have been paying more attention so its unlikely autopilot is at fault.

 

For drivers, there should not be any added liability, compared to those not using AP. For one, AP isn't legally self-driving, so the liability split is technically already 100% human driver / 0% machine AP. For another, "under investigation" normally means no legal action has (yet) be taken, and there is no certainty (yet) that any action will ever be taken. Drivers have not (yet) been given any form of legal notice beyond what is in the Owner's Manuals.

Drivers using AP and other ADAS systems properly, should have nothing to worry about. It is only those who let their expectations get ahead of current technical reality, using AP / ADAS as a crutch to do things they wouldn't do in its absence, who should have any liability concerns, along with the manufacturers who encourage or entice them to do those things.

 

That would be an additional segment of selection bias, not "the" (as in only) selection bias. You acknowledge more than one selection bias in your next paragraph, by listing some of them:

I don't believe I've suggested that AP per se, is unsafe or is killing people. It is drivers, either of their own volition or through the encouragement of marketeers or fan-bois, putting their expectations ahead of current technical reality, who might be doing so. Common human nature does push people in that direction, and any single human's lifetime experience is too small a sample base to derive an answer of "no problem". The current opaque non-independent statistics are not allowing us to see a proper answer.

 

Autopilot hardware is standard on all recent Teslas. Autosteer is all one gets in the base price. Perhaps the middle tier Autopilot has been discontinued since I last looked in December 2023 now that the full FSD (supervised) has dropped from $12,000 to $8,000 or $99/month?

At $1,200 I'd take it! If facing a long road trip perhaps I'd rent it for a month. But at current prices I can pass, Autosteer is pretty good on interstate.

Not happy with panic braking when a vehicle crosses in front of me, or when a big truck is coming around a left hand curve toward me while I'm going right in the curve. When I see this before the adaptive cruise control sees it, I put foot on accelerator to override.

 

Same here.

Yeah, FSD is more chilled than AP on those situations.

I haven't had that problem since the update of spring 2022. It would always panicked brake before then. Made using TACC and AP painful going to/coming from the cottage, especially in winter when the travelling was mostly in night time. But since then, nope.

 

Yes I guess i should have worded it as the main selection bias that I have heard and seen about autopilot is .... I did acknoledge that autopilot is only in newer vehicles (since 2012, but most were sold in the last few years) and this along with main number of miles being freeway make the numbers look better than if you coordinated with newer cars on the highway.

+1

Yes you haven't said it, but it has been repeated over and over again that teslas are dangerous and autopilot should be banned because its dangerous. The problem with the statistics is not that they are non independent, which would help, it is that we don't have comparable statistics from comparable vehicles.

 

I'm definitely not part of the ban crowd, these sorts of ADAS are a necessary step towards true self-driving. I'm more concerned about the operators who are getting over their skis, to the detriment of other road users, and reining them in. We need better gauging of the extent of this issue, before it becomes worse.

Tesla should be able to produce an apples-to-apples comparison of their own vehicles under the same or similar conditions. I.e. a comparison in which the Tesla non-AP crashes include only those miles driven in AP-suitable conditions, not weighted down by the miles and crashes that occur in places and conditions where AP is non-functional or unsuitable, as is done in the OP press release. If they have produced such a comparison internally, I haven't seen it released publicly.

Ditto for FSD.

 

Too old? That limo driver was 27. The driver local to me, who used FSD to run over the in-lane motorcyclist, is 56.

Would you still do this trip in the absence of FSD? If not, why not?

 

Like I mentioned elsewhere, FSD stopped me from t-boning someone because I was distracted by those two cars stopped in the opposite lane which is why I didn't notice that car that was crossing right in front of me (and had no right of way). So that didn't make it into the statistics because nothing happened. But, there is also that guy on Reddit yesterday that said that FSD is so good that he's zoning out. This is what can make these product dangerous, zoning out. Don't ZONE OUT! It's a Level 2 ADAS, YOU are still in control of the vehicle! It pisses me off because it's these type of drivers that will have an impact on those of use that uses it as intended. Use the advantage of not micromanaging the drive to be more aware of your surroundings, not for site seeing or day dreaming! Being occasionally distracted for whatever reason can happen to all of us and it can save yours or someone else's life but don't count on it do it 100% of the time, so stay alert!

 

This zoned-out Reddit driver illustrates precisely one of the human-factor problems of a very good ADAS system that still isn't good enough for true self driving. 'Zoning out' is typical human behavior, and is magnified when fatigued.

Then add in all the drivers who use sub-self-driving ADAS to intentionally disengage from driving.

The average-crash-rate math is such that a single very bad driver offsets not just one very good driver, but many very good drivers. E.g., the national average traffic fatality rate is slightly over 1 death per 100 million VMT (100,000,000 vehicle miles traveled). (Well, it was almost that low a decade ago, but has jumped since.). If we take one bad driver who would have a death rate of 1 death per 1 million miles, how many 'perfect' drivers would he offset? His dangerous driving would offset not just one 'perfect' driver, but rather 99 'perfect' drivers. A horrid driver who would kill one person per 100k miles, would offset 999 'perfect' drivers. While these bad and horrid drivers do tend to prematurely remove themselves from the gene pool, they also take out others and badly tilt the safety statistics.

The crash rate math of less-than-true-self-driving ADAS systems, when paired with increasing numbers of zoned-out and intentionally-disengaged drivers, could work the same way. This is why we need a better gauge of this problem, and likely better reins to control it.