Data likely shows Teslas on Autopilot crash more than rivals

[Supcom]

My understanding is ProPilot Assist was initially released on 2019 models.

So it should be a good 3 years. Quite a bit less time than Tesla, but not exactly nothing either.

Do all other maker's ADAS equipped cars report directly to the manufacturer when a crash has occurred, as do Teslas? If not, then how do we know that the numbers reported by other manufacturers are complete and accurate?

Do these other brands keep log data in the car that identifies when ADAS is being used and when a crash occurs? If not, how does the manufacturer know whether a crash is reportable? Are ADAS-equipped cars required to record this information?

If a crash occurs, are first responders required to collect information from the driver (if possible) so that accident reports would identify whether the car was using ADAS near the time of the accident?

If an owner has body work done on his car, is the body shop required to access any ADAS logs and report ADAS usage to the manufacturer?

I suspect that the ADAS reporting requirements apply only to that data that the manufacturer has in their possession and would not impose a burden on them to design features in the car for that purpose. Nor would it likely impose a burden on anyone else to report any such data to the manufacturer. So, if a car is crashed with no significant injuries and taken to a third party repair shop, how does the manufacturer report this?

Tesla, by virtue of their well-connected infrastructure, does get data from any car that crashes, so long as the car is still capable of doing so. This almost certainly slants raw data toward higher numbers of accidents.

 

[TunaBug]

Not giving the NHTSA nearly enough credit, these are people who specifically work in Auto safety and who surely know more about the relevant metrics than us observers who don’t work full time jobs in this very specialized sector

I don't think that I wrote something that implied anything negative about the expertise of the people at the NHTSA. I didn't even infer that from the linked article.

Cumulative miles travelled with the system
engaged is one of many figures requested last year by the NHTSA, you can find an exhaustive list by reading through their data request template

https://static.nhtsa.gov/odi/inv/2021/INIM-PE21020-85284.pdf

They also requested logs per vehicle of hardware configurations, any changes, software versions and any changes, etc etc etc

Holy cow! Thanks for pointing that out. They didn't just ask for miles in which a Level 2 system was engaged: they asked them to provide that for each individual VIN.

I don't suppose there is anything available about the responses? I am wondering if such data is even collected by each car manufacturer?

 

[Supcom]

I don't think that I wrote something that implied anything negative about the expertise of the people at the NHTSA. I didn't even infer that from the linked article.


Holy cow! Thanks for pointing that out. They didn't just ask for miles in which a Level 2 system was engaged: they asked them to provide that for each individual VIN.

I don't suppose there is anything available about the responses? I am wondering if such data is even collected by each car manufacturer?

I would expect that, for most vehicles, any such data is available only if the car comes into their hands. If the car is taken to Joe's Body Shop, this data is not collected and so is not reported.

If the manufacturer is aware of a crash, some info may come from the manufacturer's records, like what features were sold with the car and the last software version installed by the manufacturer. But, unless there is a federal requirement to record things like ADAS usage, the manufacturer may not be able to determine whether ADAS was in use within 30 seconds of a crash, even when the car is in their possession.

I wonder what percentage of cars from each brand have crash repairs performed at the manufacturer or dealer's facilities. I expect that a higher percentage of Teslas are brought to Tesla for repairs than Hondas going to a Honda dealer.

 

[Jcl007]

I would expect that, for most vehicles, any such data is available only if the car comes into their hands. If the car is taken to Joe's Body Shop, this data is not collected and so is not reported.

If the manufacturer is aware of a crash, some info may come from the manufacturer's records, like what features were sold with the car and the last software version installed by the manufacturer. But, unless there is a federal requirement to record things like ADAS usage, the manufacturer may not be able to determine whether ADAS was in use within 30 seconds of a crash, even when the car is in their possession.

I wonder what percentage of cars from each brand have crash repairs performed at the manufacturer or dealer's facilities. I expect that a higher percentage of Teslas are brought to Tesla for repairs than Hondas going to a Honda dealer.

Yeah, a lot of pressure is put on Tesla here, and I’m not saying there should l not be an investigation, because the system does have some problems like with phantom braking. But, my previous Honda would do that to sometimes and I’ve had it try to emergency break hard when not needed, almost causing me to be rear ended.

 

[S4WRXTTCS]

Again, I go back to my earlier statement- Adaptive Cruise/TACC, AutoPilot and FSD all fall under the 'Level 2' category but are dramatically different. They have different abilities, different use cases, different expectations and different potentials for abuse. As others have said, you also need to look at accident rates per mile driven. The above numbers make a nice clickbait headline but are really impossible to interpret or compare without further details.

True, and this actually makes the other manufactures look worse than they might really be.

The reason for this is a lot of L2 vehicles aren't really L2 in a usability kind of perspective. But, they still have useful features like adaptive cruise control. I'm a strong believer in Adaptive Cruise control and I can't see why anyone who's tried it wouldn't use it. When used correctly it can reduce accidents on the freeway by increasing following distances to what they should be.

Now it's not going to prevents accidents like cell phone users swerving into a persons lane, and that kind of thing. So I can't see having hundreds of thousands of them on the road for years without crashes even if the full L2 feature was hardly ever used.

In terms of overall safety I expect TACC (adaptive cruise control) > AP (or a true equivalent) -> Manual Driving

I also expect vehicles with active safety (FCW, AEB, etc) to be significantly safer than cars without it. Tesla does a poor job statistically wise because they keep comparing themselves to all cars where most of them don't have active safety.

AP can't compete safety wise to TACC (with active safety) because the driver is going to be out to lunch. The reduction in situational awareness is a very real thing, and will show up statistically. With Tesla in particular this is compounded by the degree some customers abuse the system by doing things like driving under the influence or distracted driving (which to my knowledge Tesla still hasn't released a fleet wide update to fully use the in-cabin camera to prevent AP abuse on vehicles with the camera).

If the NHTSA is really serious about preventing L2 accidents they need to have manufactures agree to having all new L2 vehicles have connectivity (for data reporting), and they need to have Proper Driver Monitoring (to prevent abuse). With those two they can start to focus on the factors that lead to L2 accidents, and what technology is needed. There only about 7 years late.

 

[Supcom]

Yeah, a lot of pressure is put on Tesla here, and I’m not saying there should l not be an investigation, because the system does have some problems like with phantom braking. But, my previous Honda would do that to sometimes and I’ve had it try to emergency break hard when not needed, almost causing me to be rear ended.

Unless it causes and accident, phantom braking is not a part of this reporting. This is about ensuring that cars really are safer with ADAS systems by looking at accident statistics. This certainly should be studied. But releasing raw data without an accompanied analysis that takes into account variations in reporting will lead to headlines and impressions that are not valid. That could cause people to forego using ADAS features even if they are significantly safer.

 

[S4WRXTTCS]

Do all other maker's ADAS equipped cars report directly to the manufacturer when a crash has occurred, as do Teslas? If not, then how do we know that the numbers reported by other manufacturers are complete and accurate?

Do these other brands keep log data in the car that identifies when ADAS is being used and when a crash occurs? If not, how does the manufacturer know whether a crash is reportable? Are ADAS-equipped cars required to record this information?

If a crash occurs, are first responders required to collect information from the driver (if possible) so that accident reports would identify whether the car was using ADAS near the time of the accident?

If an owner has body work done on his car, is the body shop required to access any ADAS logs and report ADAS usage to the manufacturer?

I suspect that the ADAS reporting requirements apply only to that data that the manufacturer has in their possession and would not impose a burden on them to design features in the car for that purpose. Nor would it likely impose a burden on anyone else to report any such data to the manufacturer. So, if a car is crashed with no significant injuries and taken to a third party repair shop, how does the manufacturer report this?

Tesla, by virtue of their well-connected infrastructure, does get data from any car that crashes, so long as the car is still capable of doing so. This almost certainly slants raw data toward higher numbers of accidents.

My initial post covered most of those same questions with the same concern you have.

The only cars I really trust to have connectivity are:

GM Supercruise (connectivity is actually required for it to work)
Blue Cruise (I think its only equipped on vehicles with connectivity)
Lucid
Rivian
VW (only on things like the ID4)

 

[S4WRXTTCS]

Unless it causes and accident, phantom braking is not a part of this reporting.

Is this correct? From my reading of this the NHTSA requires all significant accidents to be reported where any part of the L2 system was engaged.

This makes sense as an alert driver will prevent an accident from occurring regardless of who or what caused the situation from happening.

 

[NCC81701]

AP can't compete safety wise to TACC (with active safety) because the driver is going to be out to lunch. The reduction in situational awareness is a very real thing, and will show up statistically. With Tesla in particular this is compounded by the degree some customers abuse the system by doing things like driving under the influence or distracted driving (which to my knowledge Tesla still hasn't released a fleet wide update to fully use the in-cabin camera to prevent AP abuse on vehicles with the camera).

This is speculative and I beg to differ to your conclusion. This is after all what NTSHA is investigating under this case so we will see at the conclusion of this study. But I'd argue that safety is enhanced with AP provided drivers use AP properly as intended. It leaves the driving to the car while freeing up the driver to monitor the situation around them and plan decisions. I frequently drive with AP and noticed how poorly most people keep their car in their own lanes even when they are 100% in control of their cars.

The problem isn't with the technology, the problem is lack of proper awareness and expectations for the driver on how to properly use it. Blame it on the name, blame it on all the TikTok/YT videos of a Tesla "driving itself," blame it on Elon overhyping, but whatever the case maybe, there is a misconception that AP can do more than it does. I think the broader public is slowly readjusting and realigning their expectations for what L2 systems in general and Tesla AP specifically are actually capable of and once expectations and capabilities aligned, L2 driver assist like AP is undoubtedly better than simple TACC or manual driving.

 

[Dewg]

My initial post covered most of those same questions with the same concern you have.

The only cars I really trust to have connectivity are:

GM Supercruise (connectivity is actually required for it to work)
Blue Cruise (I think its only equipped on vehicles with connectivity)
Lucid
Rivian
VW (only on things like the ID4)

Perhaps NHTSA will update their rules and regulations for SAE levels and require Tesla style logs and telemetry be transmitted and stored for retrieval after an event.

 

[Daniel in SD]

The problem isn't with the technology, the problem is lack of proper awareness and expectations for the driver on how to properly use it.

Almost all collisions are caused by human error so you could say this about automobiles in general. Autopilot just decreases one type of human error while increasing another type of human error. If the goal is reducing collisions it's the net effect that matters.

 

[EVNow]

From July 20th, 2021, to May 21st, 2022, there were 273 crashes involving Tesla vehicles using Autopilot, according to the report. The EV company’s crashes represent the bulk of the total 392 crashes reported during that period.

Other automakers didn’t come close to Tesla’s number of reported crashes. Honda, which sells its ADAS features under the brand “Honda Sensing,” disclosed 90 crashes. Subaru, which packages its ADAS under “EyeSight,” reported 10 crashes. Ford disclosed five crashes, Toyota reported four crashes, BMW reported three crashes, and General Motors, maker of Super Cruise, only disclosed two crashes. Aptiv, Hyundai, Lucid, Porsche, and Volkswagen each reported one crash.

Again BS.

Absolute numbers or per 1000 cars is an absolute BS way of calculating. Other companies do not have worthwhile AP that people use a lot.

Tell us by miles of actual usage. AP is 4 million miles per crash. Much better than general auto fleet.

 

[Daniel in SD]

Again BS.

Absolute numbers or per 1000 cars is an absolute BS way of calculating. Other companies do not have worthwhile AP that people use a lot.

Tell us by miles of actual usage. AP is 4 million miles per crash. Much better than general auto fleet.

While I agree that absolute numbers per 1000 cars is a useless metric I haven't seen the stats for airbag deployments in the general auto fleet that would be needed to compare to Tesla's 4 million mile number. And of course you would also have to correct for driver demographics, location, time of day, etc. to get a meaningful comparison.

 

[Mardak]

Here's the NHTSA press release for the report: NHTSA Releases Initial Data on Safety Performance of Advanced Vehicle Technologies | NHTSA

It links to two pdfs:

• Summary Report: Standing General Order on Crash Reporting for Level 2 Advanced Driver Assistance Systems
• Summary Report: Standing General Order on Crash Reporting for Automated Driving Systems

Just looking at absolute numbers (), California is the "most dangerous" state for any level driving system!

 

[EVNow]

Here is the main thing.

While I agree that absolute numbers per 1000 cars is a useless metric ...

Just stop there. Read the disclaimers.

Summary Incident Report Data Are Not Normalized Reporting entities are not required to submit information regarding the number of vehicles they have manufactured, the number of vehicles they are operating, or the distances traveled by those vehicles. Data required to contextualize the incident rates are limited. Data regarding the number of crashes reported for any given manufacturer or operator, therefore, have not been normalized or adjusted by any measure of exposure, including the operational driving domain or vehicle miles traveled. For example, a reporting entity could report an absolute number of crashes that is higher than another reporting entity but operate a higher number of vehicles for many more miles.​

 

[TunaBug]

Here's the NHTSA press release for the report: NHTSA Releases Initial Data on Safety Performance of Advanced Vehicle Technologies | NHTSA

It links to two pdfs:

• Summary Report: Standing General Order on Crash Reporting for Level 2 Advanced Driver Assistance Systems
• Summary Report: Standing General Order on Crash Reporting for Automated Driving Systems

Just looking at absolute numbers (), California is the "most dangerous" state for any level driving system!

[snip]

Thank you for posting links to sources. I encourage people to look at it. And to notice that at the bottom there is contact information.

While I think that the metrics posted are bogus this is perhaps a good start. Not because it's good, but because they have to start somewhere. If they can push the industry to collect data that can then be gathered and used to produce useful metrics then we would all benefit.

 

[EVNow]

This whole thing is just another FUD by people like @2101Guy

A lot of clueless people (esp. TSLAQ) have been trying to create FUD around anything & everything Tesla/Elon for years and years. Nothing new here.

FUD is so high, Elon even made a joke about it.

But remember the reality ...

 

[ZenRockGarden]

Level 2 ADAS Crashes by Reporting Entity​

Tesla

273
Honda

90
Subaru

10
Ford

5
Toyota

4
BMW

3
General Motors

2
APTIV

1
Hyundai

1
Lucid Motors

1
Porsche

1
Volkswagen

1
Source: National Highway Traffic Safety Administration

Right, and publishing the raw numbers this way without any attempt at normalizing by crashes-per-miles-driven reeks of bad-science

 

[EVNow]

Right, and publishing the raw numbers this way without any attempt at normalizing by crashes-per-miles-driven reeks of bad-science

Not just bad science.

It is a deliberate attempt at FUD. I won't be surprised if OP has some undisclosed conflict of interest.

 

[sleepydoc]

Here is the main thing.

Just stop there. Read the disclaimers.

Summary Incident Report Data Are Not Normalized Reporting entities are not required to submit information regarding the number of vehicles they have manufactured, the number of vehicles they are operating, or the distances traveled by those vehicles. Data required to contextualize the incident rates are limited. Data regarding the number of crashes reported for any given manufacturer or operator, therefore, have not been normalized or adjusted by any measure of exposure, including the operational driving domain or vehicle miles traveled. For example, a reporting entity could report an absolute number of crashes that is higher than another reporting entity but operate a higher number of vehicles for many more miles.​

Wait - are you telling me @2101Guy made a post with a dramatic, click bait headline about something reflecting poorly on Tesla that lacked context or nuance and omitted pertinent details? I am shocked. SHOCKED, I tell you!