My friend's model X crashed using AP yesterday

[MP3Mike]

I've said it before and will say it again: the only proper methodological approach is: compare AP equipped otherwise identical cars (battery/performance trim) with and without AP over all miles driven (with and without AP enabled) and you'll get a "true" (real world) measure of if AP equppied cars are safer. Sort of like an "intention to treat" analysis if that means anything to anyone.

Not really, because you still have different drivers driving in different road conditions. What if all of your "good" drivers are in your non-AP stats, while all the reckless drivers are in your AP stats?

 

[omarsultan]

#sigh If this follows the battery fires, discussion on whether n=3 is a valid sample size in 3...2...1...

 

[Johan]

Not really, because you still have different drivers driving in different road conditions. What if all of your "good" drivers are in your non-AP stats, while all the reckless drivers are in your AP stats?

Then that's part of your real world analysis. That's the beauty of my methodology: you'll never over estimate the value of AP.

 

[xkwizit]

Privacy is a very legitimate concern indeed.

When you buy a Tesla, you have got to understood that your car is logged 24/7.

The best way to keep your privacy is the absence of logging. No logging means no one can reconstruct your detailed timeline.

Thus, I have no answer when you are a Tesla owner because those cars have all kinds of sensors and can log minute details as shown in Top Gear Roadster incidence, New York Times Model S reporter Broder, first Model S lemon lawsuit that logged the frunk was opened with fuse box tampered...

@Tam - yes with connected devices and apps like facebook there is no concept of total privacy.

However, that is not to be confused with the scrutiny that every Model X owner goes through by media and other places if they happen to be involved in an accident. For example this guy from Seattle who crashed his car in Montana. For whatever reasons he misunderstood that he could drive hands free on a rural curved road in a remote area of the country. I am sure there are thousands others who have done those kind of stupid mistakes while driving. But their life and every word they utter does not get the scrutiny that this person is going through. It's just another vehicle crash from which everyone in the car walked away. So many others like this happen every day. Leave it there media and other obsessed people

 

[JohnQ]

Two billion miles driven comes from Tesla's published numbers. Only one non-traffic fatality has occurred in those two billion miles driven. For Autopilot, it's one fatality per 130 million miles.

You can't compare AutoPilot to U.S. fatality rates, as Teslas are much more advanced, safety-wise, than the general U.S. car population. Much larger crumple zones, automatic emergency braking, strong aluminum shell, more airbags, etc. You must compare Tesla to Tesla.

There's a great binomial confidence calculator online here: Epi Tools - Calculate confidence limits for a sample proportion. Enter in your own assumptions and confidence levels and see what kind of confidence intervals you get.

Examples:
http://epitools.ausvet.com.au/conte...pleSize=130000&Positive=1&Conf=0.85&Digits=12

and:

http://epitools.ausvet.com.au/conte...leSize=2000000&Positive=1&Conf=0.85&Digits=12

You have to pick the number of trials--that is, the number of times a human driver or Autopilot could have produced a fatality, but didn't. Bad weather, merges, debris, traffic, construction, idiots, poor road markings, etc. Above, I've picked one trial for every thousand miles. Above, the confidence intervals don't intersect, but it's only at a CI of 0.85. I was bummed about that, but then realized that I'm fine with it for practical decision making.

Without quibbling about number of trials, what is the validity when the probability of "success" in each test varies?

Apologies to others regarding this statistics digression ... er, deviation

 

[KaiserSoze]

This stat is completely wrong. There are way more than just one Tesla fatality just from the reported ones (not including ones that may not be reported in the media or anywhere on the internet).

I found three (non autopilot ones) just from the first page of google.
Google

Here's another one reported on TMC that did not get covered by the press as a Tesla fatality:
New Tesla Fatality - Single Car Accident.

I tried to include only "traffic" fatalities: no suicides or car thieves. Roadsters also excluded, on the grounds that they're significantly different from the S and X (cars with Autopilot). Fortunately so few were made (2,450) that they don't affect total miles driven.

 

[xkwizit]

I tried to include only "traffic" fatalities: no suicides or car thieves. Roadsters also excluded, on the grounds that they're significantly different from the S and X (cars with Autopilot). Fortunately so few were made (2,450) that they don't affect total miles driven.

@KaiserSoze - so maybe you're looking at a purely statistical approach and you likely have your maths correct. But numbers don't tell the whole story. And almost 99.99% of users who use AP regularly will vouch that AP makes for safer driving experience. And no my 99.99% is not based on any mathematical algorithm or factual data. It's just a number to represent a majority of users.

 

[Johan]

@KaiserSoze - so maybe you're looking at a purely statistical approach and you likely have your maths correct. But numbers don't tell the whole story. And almost 99.99% of users who use AP regularly will vouch that AP makes for safer driving experience. And no my 99.99% is not based on any mathematical algorithm or factual data. It's just a number to represent a majority of users.

You need to read up on common biases with regard to statistics. To mention the most important ones in this case:
- selection bias
- recall bias
- reporting bias

 

[electricity]

@KaiserSoze - so maybe you're looking at a purely statistical approach and you likely have your maths correct. But numbers don't tell the whole story. And almost 99.99% of users who use AP regularly will vouch that AP makes for safer driving experience. And no my 99.99% is not based on any mathematical algorithm or factual data. It's just a number to represent a majority of users.

It definetly makes it safer for other people to. I don't think I ever driven in my lane and right on the speed limit ALL THE TIME like I do now with AP

 

[SR22pilot]

Two billion miles driven comes from Tesla's published numbers. Only one non-traffic fatality has occurred in those two billion miles driven. For Autopilot, it's one fatality per 130 million miles.

Even the best statistical analysis is useless if the initial dataset is flawed. So when a car goes off a cliff where it wouldn't have if on AP you rule it a suicide and dismiss it? Heck I could speculate that the AP was a suicide since the driver did nothing. However, I haven't and I am not speculating. Clearly you have gone to some extreme measures to exclude fatalities. I could go the other way. How about accidents where people in other cars were killed that wouldn't have happened had the Tesla been on AP?

 

[SW2Fiddler]

They only comment about accidents when either owners or the public say something negatively that is contrary to the cars' safe performance.

If you don't like this kind of potential exposure, then Tesla is not for you.

That is a bit extreme. I am still going to buy a Tesla. I own Tesla stock. I am just not a Kool-Aid drinker and think they could use some criticism for some of their policies.

Some might say that comparing "those other" Tesla and TSLA owners to a religious suicide cult is ...a bit extreme as well. Possibly? Maybe?

 

[Az_Rael]

Some might say that comparing "those other" Tesla and TSLA owners to a religious suicide cult is ...a bit extreme as well. Possibly? Maybe?

Alright, you got me.

 

[stopcrazypp]

I tried to include only "traffic" fatalities: no suicides or car thieves. Roadsters also excluded, on the grounds that they're significantly different from the S and X (cars with Autopilot). Fortunately so few were made (2,450) that they don't affect total miles driven.

That is plain old cherry picking. For the cliff incidents (there were two separate ones actually) it was never conclusive that they were suicides. In fact, I remember it was said that if autosteer was available at the time, perhaps those wouldn't have happened.

In that thread I linked, a member who asked the neighbors got a response that it was an Model S.
New Tesla Fatality - Single Car Accident.
New Tesla Fatality - Single Car Accident.

 

[James Anders]

What also complicates this entire thing is the human factors/user interface aspect.

I was thinking more about the tragic Anton Yeltsin death with the Jeep. In that case, even though he was familiar with how to properly put the car into part - something obviously happened to make the park action less obvious.

No, it's not a direct correlation to AP problems but it does point out that these things aren't always black and white. There can be very subtle design choices that may not significantly change the physical operation from a machine perspective but certain minor design changes can have a significant impact over large user populations.

 

[xkwizit]

Now we can clarify some of the question in this thread, the autopilot was NOT engaged when the crash happened according to Teslas logs.



Autopilot wasn’t on during Model X crash in PA and Musk says it would have prevented accident

That's a different thread. This one is Montana crash.

 

[vigge50]

That's a different thread. This one is Montana crash.

Sorry, didn't read well enough

 

[CuriousG]

Here's a reenactment of the latest AutoPilot crash. I don't take credit for this but saw it posted somewhere else.

http://imgur.com/C2Gde3j.gif

 

[xkwizit]

Here's a reenactment of the latest AutoPilot crash. I don't take credit for this but saw it posted somewhere else.

http://imgur.com/C2Gde3j.gif

@CuriousG - that's so funny. Reminds me of another one some one posted of kids driving over each other.

 

[Joeski1]

I don't see why TESLA can't offer refunds for folks that want out of AP... I can't find alotta use for it personally.. shame I was misled at the retailer... had I known it was of such limited use in my daily driving.. I would not have spent the money on it...

don't really care about much else here... two near accidents and off the road while using it.. and I'm done... the system is too limited to be of use except on interstate highways and turnpikes perhaps... but around town.. and on lazy back-roads.. esp. ones with curves and hills.. it's not good enough... and not recommended either....

they need to put a large print disclaimer on their site to warn folks it's really designed for high speed , limited cross traffic highway use only...a few test drives and a 30 hour loaner is NOT enough time behind wheel time to make a large cost decision about this feature

 

[MP3Mike]

I don't see why TESLA can't offer refunds for folks that want out of AP... I can't find alotta use for it personally.. shame I was misled at the retailer... had I known it was of such limited use in my daily driving.. I would not have spent the money on it...

don't really care about much else here... two near accidents and off the road while using it.. and I'm done... the system is too limited to be of use except on interstate highways and turnpikes perhaps... but around town.. and on lazy back-roads.. esp. ones with curves and hills.. it's not good enough... and not recommended either....

they need to put a large print disclaimer on their site to warn folks it's really designed for high speed , limited cross traffic highway use only...a few test drives and a 30 hour loaner is NOT enough time behind wheel time to make a large cost decision about this feature

Really, you would give up TACC and the auto parking features too? (Remember it is a bundle, not à La Carte.) I think for most people TACC alone is worth it. But maybe you don't drive in rush hour traffic on the freeway much.