Another tragic fatality with a semi in Florida. This time a Model 3

[acoste]

You are welcome to the opinion their analysis was flawed. They don't think it was.

Further, "I think their analysis was flawed so I'm gonna toss over 90% of their data and draw a totally different conclusion on the little that is left" is what the other guy did- which is.... pretty questionable itself.

If you wanna say you don't feel sure either way I guess you can- but there's at least as many problems with that one dudes counter-analysis, if not more, than with the original one.

YM
Based on an analysis a single guy being paid by SOMEBODY to chase after tesla for a few years now tells you is wrong
HTH!

I already went into some detail on why that articles analysis is crap...you appear not to have noticed?

Again. This is just talking. You don't back up your claims. You say he tossed 90% of the data. No he didn't. Check the data. Make your own conclusions.

NHTSA was unwilling to release the data. Once the report I linked got published by many sources NHTSA didn't sue anyone.

For one they just used "news stories" to guess at the # for Tesla worldwide (vs NOT doing that for "all luxury cars") so they're comparing Apples and Candy Bars there.


2 problems here...

1) Per billion miles driven is a lot more useful stat than per million vehicle years. Teslas tend to get driven a lot more often than a Bugati that would also be a "luxury" car. A car sitting among a dozen others in a rich dudes garage and never going anywhere isn't likely to kill anybody. So maybe an analysis that counts those equally with a car driven 50k miles a year is a crap analysis?

2) You admit you're guessing on the # of fatal tesla accidents-unsure if AP was even on for one of them and again 2 of THOSE were when AP was used someplace it's explicitly NOT supposed to be used so counting those isn't reasonable either.

1) If your turn this metric to miles driven / accident you can't compare anything because there is no data on how many fatal accidents happened on roads where Autopilot can be turned on.

2) 3 of those fatalities are with AP on confirmed. Last (4th) one is pending. And my number is an underestimation. Calculates all type of fatal accidents for the luxury cars (even killing a bicyclist), but Tesla's AP is used on specific roads only and not all the time.
Your Bugatti claim has no merit. Makes very little difference. Look at what cars are in the luxury bucket.
If you are still blaming the driver, read the 3rd article again. Yes he was at fault but this is human behavior. Do you think the Apple engineer was stupid and didn't know about the capabilities of the AP?

 

[Knightshade]

Please reread the report. It said "Driver's hands were not detected". Since you apparently don't have a Tesla, I shot a video the other day for the people like you.

I do have one actually- it's why I know how useless your video is.

It's 4+ minutes of driving in a straight line.

AP detects hands on wheel with a torque sensor- gauging resistance to AP turning the wheel.

Which it's not doing in a straight line.

But is doing when not going straight.

The crash in question the car wasn't going straight so it would've detected torque against the wheel if there'd been any.


Here's what's known- he had received several visual, and one auditory, warning about driving without hands on wheel previous in the drive, and his hands were not detected on the wheel for at least 6 seconds prior to the crash (where again the car was in the middle of a non-straight-line manouver)

We also know the driver had:

about five seconds and 150 meters of unobstructed view of the concrete divider with the crushed crash attenuator, but the vehicle logs show that no action was taken.

So he was clearly not paying any attention to where the car was going, as AP specifically instructs the driver to do.

Again. This is just talking. You don't back up your claims. You say he tossed 90% of the data. No he didn't. Check the data. Make your own conclusions.

To reach his conclusion he used only 5,714 vehicles of the total 43,781 vehicles.

Which is tossing pretty close to 90% of the total dataset. 87% if you're REALLY butthurt about rounding a number though.

So no, I'm not "just talking" and no I'm not "backing up" my claims.

Here BTW was my original claim on the dataset thing:

The first one for example simply tosses out almost 90% of the NHTSA data

87% is indeed almost 90%. I do seem to have reversed the phrasing in a later post, apologies for that- but if the best you've got is quibbling over "Needed to throw out nearly 90% of the data to reverse the conclusions" versus "needed to throw out slightly more than 90% of the data to reverse the conclusions" you haven't a very strong argument do you?

1) If your turn this metric to miles driven / accident you can't compare anything because there is no data on how many fatal accidents happened on roads where Autopilot can be turned on.

... I'm not even sure what this statement means? AP can be engaged nearly anywhere- it's just not intended for use other than on divided highways. We generally know when AP was engaged in a fatal accident- it's in the logs of the car. (there's one exception where the car was so badly destroyed the logs were not available)


Certainly miles driven is a much more useful metric than "car years" though since car years counts cars never actually driven, which is nonsensical.

2) 3 of those fatalities are with AP on confirmed. Last (4th) one is pending.

And 2 of them on roads where AP is explicitly not intended to be used.

So 4 becomes 2 if your goal is still to "blame" AP instead of the idiot behind the wheel.

And my number is an underestimation.

This is called assuming facts not in evidence.

Calculates all type of fatal accidents for the luxury cars (even killing a bicyclist),.

But then also counts cars that just sit in garages and never go anywhere- which is why it's an insanely bad metric for this.

Your Bugatti claim has no merit. Makes very little difference. Look at what cars are in the luxury bucket.

Here's you, again, doing exactly the thing you accuse me of- making claims without support or evidence.

The fact you're using a metric that counts cars that aren't even driving anywhere makes it clear it's a dumb metric compared to using miles driven

If you are still blaming the driver, read the 3rd article again. Yes he was at fault but this is human behavior.

I'm glad you finally admit it- but makes it weird you're still trying to blame the car for what was human error.

 

[verygreen]

It's 4+ minutes of driving in a straight line.

want me to upload another version wher the road is curvy with the exact same result? Did you miss me wiggling the wheel some?

he had received several visual, and one auditory, warning about driving without hands

No! he received some warnings about his hands not being detected! Those are different things. I get the warnings (= nags) all the time (not the red beeping one, though even that is easy to get lately)

So he was clearly not paying any attention to where the car was going, as AP specifically instructs the driver to do.

Did you miss Wk's post where he details how when a car wants to do a sudden maneuver on AP2, it does it with so much force, unless you hold the steering wheel super tightly - it will overpower you? Esp if it caught you by surprise?

The car was moving at a high speed at a familiar location, you lightly hold the steering wheel because I know all is fine (hence the periodic nags).
Suddenly the car jerks with a great force and overpowers you (due to surprise and force). I had a similar experience at 75mph on a country road (back when I tried to experiment with AP more there) where the car tried to steer onto oncoming lane to hit an oncoming truck. It does do it with a significant force amplified by speed. Unless you have an iron grip - easy to lose it! Even then it's scary. I am glad it never happened to you, but it does happen, there are multiple reports.

There's no viedo feed to know for sure, but if the movement was sudden there might have been no time to react.

I am not saying this is exactly what happened, what I am saying is it's way too easy to read what Tesla said in a particular way (remember NHTSA report has no been released yet! I wonder why and when it'll be out) where it's clear all they are interested in is to shift the blame away and engineer their reports accordingly.

 

[Knightshade]

Did you miss Wk's post where he details how when a car wants to do a sudden maneuver on AP2, it does it with so much force, unless you hold the steering wheel super tightly - it will overpower you? Esp if it caught you by surprise?

AP can not overpower the driver. Just like the throttle can't overpower the brakes.

Certainly if you're holding the wheel lightly it might move the wheel in a way you're not expecting. But you'd realize that was happening very quickly if you're actually holding said wheel and actually paying attention. Which he wasn't.

That's the great thing about the logs.

It tells us the car was headed for the barrier it hit for at least 5 full seconds, and 150 meters, where the driver would've had a clear view it was heading toward the barrier while the driver did nothing about it.

He wasn't paying attention. Or he'd have tried to avoid the barrier.


Here BTW is a video of someone driving in the same place- and the car behaves the exact same way... and since the driver is paying attention he has PLENTY of time to correct for it.

Tesla on autopilot nearly crashes in exact same spot Model X crashed | Daily Mail Online

I had a similar experience at 75mph on a country road (back when I tried to experiment with AP more there) where the car tried to steer onto oncoming lane to hit an oncoming truck.

So you were using AP in an area it's explicitly not intended to be used then- a non-divided highway with opposing/on-coming traffic.

You might not be the right person to be discussing how to correctly and safely use AP

 

[dmd2005]

Driver dead after Tesla crashes into pole on Lougheed Highway

Another fatality involving a Model X and sounds like speed was a factor and there was some thermal runaway with the batteries catching fire.

 

[Dangerous Fish]

Driver dead after Tesla crashes into pole on Lougheed Highway

Another fatality involving a Model 3 and sounds like speed was a factor and there was some thermal runaway with the batteries catching fire.

Looks like an X to me

 

[StellarRat]

I just don't see a case against Tesla for using Autopilot and getting yourself killed. Unless it actually wrested control from the driver and crashed the car there is absolutely no way you can blame it or Tesla for wrecks while it was engaged. The driver has total control of car any time he/she needs it and is told (and monitored) repeatedly to maintain vigilance while driving even with AP engaged. How is this any different than blaming your cruise control if you get in a wreck while it's turned "on"? That's also automation.

Show me a case where the driver was paying attention and EAP wrecked the car somehow and we'll talk.

 

[chronopc]

Driver dead after Tesla crashes into pole on Lougheed Highway

Another fatality involving a Model 3 and sounds like speed was a factor and there was some thermal runaway with the batteries catching fire.

I don't believe that's a Model 3. Looks like there's a spoiler so it looks like an X?

Has there been a report of a Model 3 catching on fire?

 

[verygreen]

That's the great thing about the logs.

It tells us the car was headed for the barrier it hit for at least 5 full seconds, and 150 meters, where the driver would've had a clear view it was heading toward the barrier while the driver did nothing about it.

I'll hold my judgment here until NHTSA releases their report. Tesla is obviously a very biased party here.

So you were using AP in an area it's explicitly not intended to be used then- a non-divided highway with opposing/on-coming traffic.

You might not be the right person to be discussing how to correctly and safely use AP

Well, like I said I was experimenting and with extra full attention and it still caught me by surprise and somewhat scared. If you are not waiting for it, it totally can make the car jerk noticeably at high speed and it would be hard to catch.

And as I said in many other places - I no longer use AP outside the highways (and even there - very sparingly) because I do find it somewhat unsafe and also not super convenient with a bunch of nags of increasing frequency and annoyance.

 

[dmd2005]

Looks like an X to me

My bad, yes it looks like an X...just editted the post.

 

[acoste]

The crash in question the car wasn't going straight so it would've detected torque against the wheel if there'd been any.
Here's what's known- he had received several visual, and one auditory, warning about driving without hands on wheel previous in the drive, and his hands were not detected on the wheel for at least 6 seconds prior to the crash (where again the car was in the middle of a non-straight-line manouver)

The car was going straight, that's exactly the problem. Should have made a slight left turn for exiting.

The last warning the driver got was like 15 minutes before the crash.

To reach his conclusion he used only 5,714 vehicles of the total 43,781 vehicles.

Which is tossing pretty close to 90% of the total dataset. 87% if you're REALLY butthurt about rounding a number though.

You keep saying that but this is totally wrong. He used all the data. And he used logistic regression for the missing data. It is up to you if you accept his calculations. But you can't miss the mistakes NHTSA did in favor of AP.





13% of data: all data provided.
> this shows 59 percent increase in the airbag deployment crash rate for AP.

34% of data: missing info on AP installation date:
> NHTSA assumed this is all AP miles even though there were 3 airbag deployments under "before AP".

20% of data: gap between last before AP and next after AP
> NHTSA skips the miles in the exposure gap.
This method is in favor of AP and unrealistic.

33% of data: no last mile before AP install reported
> NHTSA counts 0 miles for "before AP" even though they counted 15 airbag deployments for this category. Same thing as for category 2.

Conclusion:
Only the first category is correct and it's 13% of the total data. Calculations for the remaining 3 categories are flawed and in favor of AP in NHTSA's report.

Read it yourself for further details: http://www.safetyresearch.net/Library/NHTSA_Autosteer_Safety_Claim.pdf

 

[acoste]

And 2 of them on roads where AP is explicitly not intended to be used.

So 4 becomes 2 if your goal is still to "blame" AP instead of the idiot behind the wheel.

yeah, we can fake it that it didn't happen

But then also counts cars that just sit in garages and never go anywhere- which is why it's an insanely bad metric for this.

Makes very little change. If you think it's not, prove it with NHTSA's data.

The fact you're using a metric that counts cars that aren't even driving anywhere makes it clear it's a dumb metric compared to using miles driven

I agree if conditions are the same (meaning calculating the same type of miles for all cars where AP is in use) otherwise it is meaningless.

And 2 of them on roads where AP is explicitly not intended to be used.

..

I'm glad you finally admit it- but makes it weird you're still trying to blame the car for what was human error.

This is what shows you don't understand human behavior.

My first blame is on the truck driver.
Second blame is on Tesla Inc (not the AP) due to reasons the 3rd article mentions.
Third blame is on the Tesla driver.

 

[bhzmark]

The charts are nice. But they still do not explain why you have to throw out all the mileage and airbag deployments for cars known to have AP and known not to have AP.

Adding up the miles from cars with AP yields 235,970,173 with 187 airbag deployments yielding an airbag deployment every 1.2 million miles.

While miles from cars without AP yields 64,672,074 with 68 deployments yielding an airbag deployment for every .951 million miles. Showing AP has a slightly better deployment record even under that earliest implementation of AP which has been much improved since then.

The car was going straight, that's exactly the problem. Should have made a slight left turn for exiting.

The last warning the driver got was like 15 minutes before the crash.




You keep saying that but this is totally wrong. He used all the data. And he used logistic regression for the missing data. It is up to you if you accept his calculations. But you can't miss the mistakes NHTSA did in favor of AP.





13% of data: all data provided.
> this shows 59 percent increase in the airbag deployment crash rate for AP.

View attachment 387804

34% of data: missing info on AP installation date:
> NHTSA assumed this is all AP miles even though there were 3 airbag deployments under "before AP".

View attachment 387805


20% of data: gap between last before AP and next after AP
> NHTSA skips the miles in the exposure gap.
This method is in favor of AP and unrealistic.

View attachment 387806

33% of data: no last mile before AP install reported
> NHTSA counts 0 miles for "before AP" even though they counted 15 airbag deployments for this category. Same thing as for category 2.

View attachment 387807


Conclusion:
Only the first category is correct and it's 13% of the total data. Calculations for the remaining 3 categories are flawed and in favor of AP in NHTSA's report.

Read it yourself for further details: http://www.safetyresearch.net/Library/NHTSA_Autosteer_Safety_Claim.pdf

 

[acoste]

The charts are nice. But they still do not explain why you have to throw out all the mileage and airbag deployments for cars known to have AP and known not to have AP.

Adding up the miles from cars with AP yields 235,970,173 with 187 airbag deployments yielding an airbag deployment every 1.2 million miles.

While miles from cars without AP yields 64,672,074 with 68 deployments yielding an airbag deployment for every .951 million miles. Showing AP has a slightly better deployment record even under that earliest implementation of AP which has been much improved since then.

You are doing the same mistakes what NHTSA did.

You added all the 2nd category miles under AP while that is not the case. There were airbag deployments reported as nonAP in that category which shows one can't count all the miles to AP.
You didn't add any miles to nonAP here.

In the third category there is a huge gap. Some belongs to AP, some to nonAP. You skipped those numbers just like NHTSA.

And in the 4th you didn't add any miles to nonAP here however there were 15 airbag deployments there.



EDIT: I see now you are skipping the unproven data. That isn't an accurate method either.
In the 2nd category it adds more miles to the AP than reality.
In the 3rd category it skips the information that the cars didn't have airbag deployment during the gap miles.
Correct equation for cat 3 looks like:
(22.7M miles + miles of AP cars in their nonAP gap miles) / 36
(63.2M miles + miles of AP cars in their AP gap miles) / 45
using 22.7M / 36 and 63.2M/45 is an overestimation for both. Question is how much.

 

[Knightshade]

The car was going straight, that's exactly the problem.

I mean, now you're just outright lying, so I think we are done here.

https://www.ntsb.gov/investigations/AccidentReports/Reports/HWY18FH011-preliminary.pdf


NTSB report on the crash analysis (bold added to point out where no, he wasn't going straight when AP turned the car to the gore area.

As the Tesla approached the paved gore area dividing the main travel
lanes of US-101 from the SH-85 exit ramp, it moved to the left and entered the gore area

Further in the analysis-

At 7 seconds prior to the crash, the Tesla began a left steering movement

(coincidentally, his hands were not detected on the wheel at all after this point... which would not be true if he'd been paying any attention and had tried to avoid the accident by turning the wheel or resisting the left steering movement AP had initiated)

I will add one other comment about dude who decided he only liked 13% of the total NHTSA data....specifically when you remark:

It is up to you if you accept his calculations

there's very little out there about his motivations, but one does find ONE reference to something he's done before...

Before Toyota, There Was Ford-Firestone - FairWarning

The relevant bit?

The report comes from statisticians Randy and Alice Whitfield of Quality Control Systems Corp., a Maryland firm that does research for plaintiff lawyers and other clients

So basically him and his wife are the entire company far as I can tell, and they prepare paid "analysis" for ambulance chasers.

Can't imaging why anybody'd want to question their calculations.....



Lastly as a general point- the data is largely AP1 data- so not super relevant to Model 3 anyway. Tesla publishes their own, much more recent data (with a ton more miles a year driven) again showing AP is a ton safer than non-AP driving... but of course those who don't want to believe their numbers won't.

 

[acoste]

I mean, now you're just outright lying, so I think we are done here.

https://www.ntsb.gov/investigations/AccidentReports/Reports/HWY18FH011-preliminary.pdf


NTSB report on the crash analysis (bold added to point out where no, he wasn't going straight when AP turned the car to the gore area.

Thanks for sharing. I haven't seen this report yet. I live in this area and I barely need to move the steering wheel when going in either directions because the angle is so little. It's not like a curve, it needs just as much effort as keeping a car in lane.

Earlier I thought the vehicle was in the far left lane and following the cracks in the concrete, but based on the report it was in the second left lane and following the brightest white line.

there's very little out there about his motivations, but one does find ONE reference to something he's done before...

Before Toyota, There Was Ford-Firestone - FairWarning

The relevant bit?

What's wrong here? The company (Ford) did not dispute the data cited by Quality Control Systems.

So basically him and his wife are the entire company far as I can tell, and they prepare paid "analysis" for ambulance chasers.

Can't imaging why anybody'd want to question their calculations.....

Again, just look at the data by yourself. Make your own conclusions. No need to believe them. Based on the data I don't know if they are right, they calculate probabilities. There is always a chance for outliers or the remaining 5%. But I can see that the NHTSA's analysis is flawed.

Tesla publishes their own, much more recent data (with a ton more miles a year driven) again showing AP is a ton safer than non-AP driving... but of course those who don't want to believe their numbers won't.

Comparing what to what? AP miles versus non AP miles? Makes no sense since AP mainly drives on highways as you said. While accidents happen everywhere and in any weather. (My commute is 20% local 80% hw in distance and 50-50 in time).
Sorry I'm skeptical of any data that comes out of Tesla... Just think of the recent range increase of LR RWD claiming "where we are able to improve the efficiency... " yeah right

 

[bhzmark]

I see now you are skipping the unproven data. That isn't an accurate method either.

Yes I excluded only the data that was unproven, and added up the data that was designated as AP or not AP. Sorry the results don't fit your narrative and confirmation bias.

 

[seattlite2004]

More info: Highway Preliminary Report: HWY19FH008

 

[Az_Rael]

"Preliminary data from the vehicle show that the Tesla’s Autopilot system—an advanced driver assistance system (ADAS) that provides both longitudinal and lateral control over vehicle motion—was active at the time of the crash[1]. The driver engaged the Autopilot about 10 seconds before the collision. From less than 8 seconds before the crash to the time of impact, the vehicle did not detect the driver’s hands on the steering wheel. Preliminary vehicle data show that the Tesla was traveling about 68 mph when it struck the semitrailer. Neither the preliminary data nor the videos indicate that the driver or the ADAS executed evasive maneuvers."


Bold mine. So, sadly, this sounds a lot like turning on AP because you just dropped something or your phone got a text, etc.

I personally have been guilty of doing this occasionally on what I perceive as an empty clear straight freeway. Be careful out there.

 

[Daniel in SD]

More info: Highway Preliminary Report: HWY19FH008

I wish they would stop saying that the car didn't detect that the drivers hands were on the steering wheel in these accidents. The car didn't detect any torque on the steering wheel, it has no sensors to detect whether or not the driver was holding the steering wheel.