NHTSA Finds Tesla Accident Rate Drops 40% After Autosteer Installed!

[bhzmark]

NHTSA Full Report on Tesla Fatal Autopilot Crash | Tesla Motors | Tesla Model S

Has the full report. Worth reading.


It notes reduction of airbag events in AP cars by comparing prior to October 2015 when autosteer was not enabled but most safety features were vs after AP was enabled...

 

[jbcarioca]

The man was a retired Navy seal. It's possible, though not proven, that he made a mistake that day. Assuming he did, that mistake cost him his life. I don't think he deserves to be spoken about, after death, using the language you used.

If it was not a truck, but a school bus, and 10 kids died would anything change? If not, what if the kids were family members?

I have total contempt for anybody, even a family member, who would risk driving without looking forward. I would not use a term as polite as idiot.

McRat used less temperate language than did I. I spent my own years with US military and had some pretty serious training. That may have helped me with crisis decision-making, but has not given me unique ability to watch movies and control a vehicle at the same time. Are Navy Seals somehow exempt from the penalties of inattention?

 

[Andyw2100]

McRat used less temperate language than did I. I spent my own years with US military and had some pretty serious training. That may have helped me with crisis decision-making, but has not given me unique ability to watch movies and control a vehicle at the same time. Are Navy Seals somehow exempt from the penalties of inattention?

Losing his life was a pretty big penalty. I just don't see the need to pile on.

And I'm not 100% up on all the details, but I recall at some point that a friend of his, or perhaps someone in his family said he often LISTENED to movies in the car.

None of us know exactly what happened. We know he died, we know it was not the car's fault, and in all likelihood his death was probably caused by his inattentiveness. I just don't understand why people need to say bad things about him after the fact.

Apparently I'm in the minority.

 

[EinSV]

Losing his life was a pretty big penalty. I just don't see the need to pile on.

And I'm not 100% up on all the details, but I recall at some point that a friend of his, or perhaps someone in his family said he often LISTENED to movies in the car.

None of us know exactly what happened. We know he died, we know it was not the car's fault, and in all likelihood his death was probably caused by his inattentiveness. I just don't understand why people need to say bad things about him after the fact.

Apparently I'm in the minority.

I was disappointed, but not surprised, that the NHTSA made no mention in the report of a critical factor that may well have led to Mr. Brown's death -- the lack of side impact protection on the truck he collided with.

This simple but very effective safety technology has been required in much of Europe for more than four decades. Recent very credible studies have suggested that 89% of fatal or serious injuries from side impact collisions with trailers could be mitigated by use of this well-known safety feature.

Yet in all the criticism of Mr. Brown and of Tesla, this almost never gets mentioned by the press, the NHTSA, Consumer Reports or anyone else.

The photos of the crash showing the impact directly into the passenger compartment vividly demonstrate the problem with the lack of side impact protection -- the point of impact avoids the crumple zone and directly impacts the passenger compartment. While we will never know for sure, seeing those photos suggested to me a strong likelihood that there could have been a different outcome if the US were not half a century behind Europe on requiring the adoption of this simple safety feature.

I posted more details a while back if anyone is interested (see below). This is a well-known and documented issue, including in NTSB communications with the NHTSA.

It is really a shame that the US is such a laggard on adopting such simple safety technology, which is on every trailer I have seen in my travels in Europe.

In the future, hopefully improved automated driving systems will be able to avoid this sort of collision entirely.

In the meantime, I continue to be surprised that more attention is not being paid to a simple, known safety measure that in this case may well have resulted in saving Mr. Brown's life-- side impact protection on the tractor-trailer he collided with.

I did a little digging and not only is side impact protection a mandatory safety requirement in the EU, but two years ago the NTSB recommended that the NHTSA adopt mandatory side impact protection requirements for new tractor-trailers. Their findings include a study that found that 89 percent of serious and fatal injuries resulting from side impacts with trailers could have been mitigated through the use of side impact protection.

http://www.ntsb.gov/safety/safety-recs/recletters/h-14-001-007.pdf (relevant bits start at page 7).

Here are a couple of excerpts from the NTSB's letter to the NHTSA:

Side underride. One reason why collisions with the sides of tractor-trailers are hazardous is that side underride may occur during these collisions.23 Both belted and unbelted occupants are vulnerable to injuries as a result of side underride. Side underride occurs when passenger vehicle bumpers are not at the same height and do not engage the substantial side structure of tractor-trailers. Side underride collisions are an important safety problem because they defeat crumple zones and prevent air bag deployment, both vital safety advances in improving protection of passenger vehicle occupants during crashes.24,25,26 Airbags will not deploy in some underride collisions when the sensors to trigger them are not contacted by vehicle structures.27 Crumple zones do not work as intended in underride collisions when relevant passenger vehicle structures fail to engage tractor-trailer structures. Furthermore, the occupant’s safety cage can be compromised when underride allows the passenger vehicle to sustain impacts at the level of the windshield and other areas above the hood. This effect can result in deaths and severe injuries due to intrusion of vehicle components. The adverse effects of underride collisions in defeating safety advances have been demonstrated in a study of belted occupants injured in passenger vehicles with good frontal crash test ratings.28 Head injuries were the most common type of serious injury in underride collisions, and underride collisions had higher rates of fatal and severe non-fatal injuries than other crash configurations. Intrusion into the passenger compartment was the most common contributing factor to injury in underride collisions.

******

In 2012, Brumbelow estimated that 530 passenger vehicle occupants died each year during 2006–2008 in two-vehicle collisions between passenger vehicles and the sides of large trucks.32 Brumbelow also did an independent analysis using LTCCS data and concluded that the most severe injuries sustained by passenger vehicle occupants were usually due to the truck side impacts rather than other events that may have occurred during the crashes (many crashes included multiple events that could result in injury). This was the case for 69 percent of 206 crashes in which a passenger vehicle collided with the side of a large truck. About a third of collisions with the sides of large trucks involved truck-tractors and about half involved semi-trailers.

*******

Brumbelow’s evaluation of the potential benefits of side underride guards strongly suggested that they would reduce injury severity. Of passenger vehicle occupants with serious to fatal injuries attributed to side impacts with semi-trailers35, 89 percent were considered injuries that could have been mitigated by side underride guards. For passenger vehicle occupants with serious to fatal injuries attributed to side impacts with truck-tractor cabs, side underride guards were considered potentially beneficial for 83 percent.3

 

[jbcarioca]

Losing his life was a pretty big penalty. I just don't see the need to pile on...

Apparently I'm in the minority.

Excuse me for reversing myself. I fear I am becoming more intolerant as I listen to so much hate and vitriol passing for political leadership recently.
I should know better. I should not have used those words. That said, were I to have done such a thing I would, were I still be alive, not expect to avoid harsh judgement.
That is no excuse for my word usage. It was intemperate and detracted from the point.

 

[brianman]

It might just mean that the safer drivers are buying the features.

 

[bhzmark]

The dataset appears to be all AP1 hardware cars that eventually had the AP enabled. The comparison is accident rate/miles driven before AP/autosteer was enabled, vs after it was enabled.

 

[EinSV]

The statistics on crashes while texting are pretty confirming of your view, especially as texting while driving is increasing. I have not seen definitive statistics comparing accident rates by use of smartphones or other mobile devices lately. The IIHS even states taht the definitive statistics are not currently available.
Q&As
Despite that the statements of accident cause from several Brazilian States by DETRAN (State Departments of Transportation) were so devastating that use of phones while driving is now illegal in Brazil, as it is in many countries.

It seems to me there is no need for much statistical evaluation on this one. Distractions, whatever the cause, increase accident risk. The famous idiot Tesla driver in Florida would still be alive if he'd not been watching a movie.

Mark Rosekind, the NHTSA Administrator, has spent his career studying fatigue and distraction in Transportation.
NHTSA Administrator Mark R. Rosekind
I may be biased because I have met him and talked with him about such issues. It is not coincidence that he and other in NTSB and NHTSA are strong advocates of developing automated driving aids and autonomous systems.
As an enthusiastic driver myself I'm less enthused. Frankly, as an aircraft pilot I prefer to hand fly. The is exactly zero question that machines make fewer errors than do people. The 'secret sauce' is in reducing the error rate and improving reliability.
So, in the age of Twitter for idiot Geezers we might as well accept it. Automation's safer.

Maybe McRat and I can drive together without automation on a private course somewhere?

A side benefit of Tesla's improvements in automated driving will almost certainly be accelerated safety gains for all of us error-prone human drivers. A great example of this is the recent video showing Tesla's Forward Collision Warning and AEB were activated even before the accident risk was visible to the driver. Watch Tesla’s Autopilot system help avoid a crash with superhuman sight

Recruiting some of the top engineering talent in the world, developing top-notch hardware and software, and enjoying the benefits of fleet learning are likely to make automated accident avoidance much more robust for human drivers as well as when all AP features are activated. And with the new radar and other capabilities in AP2, it is not difficult to imagine Tesla's AEB and FCW systems in AP2 enabled vehicles will allow a human driver to avoid potential collisions that would have occurred with other manufacturer's more primitive AEB and FCW systems.

To assess the safety gains from AP over the long run, it strikes me that the appropriate benchmark is not comparisons with Tesla's "AEB-plus" and "FCW-plus," which will enjoy continually enhanced accident capabilities that never would have been possible without the huge investment in engineering, hardware, software development and fleet learning necessary to implement AP. Perhaps a better benchmark would be the IIHS's analysis of the effectiveness of "traditional" AEB and FCW together. The IIHS predicts a 40% decrease in rear-end collisions if implemented fleetwide by 2022, but the evidence for reduction in overall accidents is much more modest -- perhaps in the range of about 10%. As I noted in another post, a recent IIHS study found that AEB plus Forward Collision Warning reduced overall accidents by only 6%, which did not even reach the level of statistical significance. http://orfe.princeton.edu/~alaink/S...s/IIHS-CicchinoEffectivenessOfCWS-Jan2016.pdf (see pages 1 and 15.)

The NHTSA's reported 40% reduction in accidents once AP was enabled is huge when compared to the relatively modest predicted gains from "traditional" AEB and FCW technology.

For those of us who enjoy both AP and "regular" driving, the safety benefits from AP technology ideally would show up equally in with full AP activated and without it. I don't think this will happen since the human driver will become the "weak link" in more and more situations, but I do believe the massive investment Tesla is making in AP will also have the side benefit of quite dramatic safety gains that help distracted and imperfect human drivers stay safe even when full AP is not activated.

 

[jbcarioca]

...but the evidence for reduction in overall accidents is much more modest -- perhaps in the range of about 10%. As I noted in another post, a recent IIHS study found that AEB plus Forward Collision Warning reduced overall accidents by only 6%, which did not even reach the level of statistical significance. http://orfe.princeton.edu/~alaink/S...s/IIHS-CicchinoEffectivenessOfCWS-Jan2016.pdf (see pages 1 and 15.)

... I do believe the massive investment Tesla is making in AP will also have the side benefit of quite dramatic safety gains that help distracted and imperfect human drivers stay safe even when full AP is not activated.

I agree. The statistical weaknesses of the NHTSA report are numerous, but most of them could not have been addressed without much more data than they had and more budget than they had. Bluntly they need larger and more diverse fleets as well as sufficient data to evaluate the relative performance of different driver characteristics.

Because their conclusion is what we want to hear and want to believe it's hard to discuss the limitations of the data without seeming to be opposing the conclusions. I am quite certain that AP-1 is safer than non, that TACC is safer than non, that traction control and ABS are safer than non, that AP-2 will end out being safer than AP-1. Still, apart from a handful of cases we simply do not have sufficient data to say "this is true".

The ones we now know are safer: ABS, traction control, TACC, seat belts, air bags (even Takata). Now we need lots more data, quickly! In the meantime I strongly favor advancing vehicle automation with as much speed as we prudently can. Tesla is doing just that.

 

[EinSV]

I agree. The statistical weaknesses of the NHTSA report are numerous, but most of them could not have been addressed without much more data than they had and more budget than they had. Bluntly they need larger and more diverse fleets as well as sufficient data to evaluate the relative performance of different driver characteristics.

Because their conclusion is what we want to hear and want to believe it's hard to discuss the limitations of the data without seeming to be opposing the conclusions. I am quite certain that AP-1 is safer than non, that TACC is safer than non, that traction control and ABS are safer than non, that AP-2 will end out being safer than AP-1. Still, apart from a handful of cases we simply do not have sufficient data to say "this is true".

The ones we now know are safer: ABS, traction control, TACC, seat belts, air bags (even Takata). Now we need lots more data, quickly! In the meantime I strongly favor advancing vehicle automation with as much speed as we prudently can. Tesla is doing just that.

It is almost always the case in real world applications that data are imperfect. The FDA has to make judgment calls all the time on whether to allow drugs to be placed on the market with limited data sets. The same is true with automotive safety tests -- I was quite surprised to read the recent IIHS report which could not find statistical signifcance in the reduction of overall accidents with AEB/FCW, despite a significant reduction in rear-end collisions.

But because information and data is necessarily imperfect, real world decisions must be made based on the best information available. We don't have access to the full data set, but the 40% reduction in accidents reported by the NHTSA after AP was enabled is a very dramatic number. This is especially so with a relatively good control group of other Tesla drivers driving similar cars (other Teslas). I suppose there could be some weird statistical quirks, but IMO the suggestions others have made (e.g., drivers who choose AP are inherently safer or more experienced) seem highly unlikely to explain such dramatic results.

I believe most people consider seat belts the most important automotive safety feature ever adopted. After almost 70 years of refinements since their introduction in 1949, the CDC estimates that seatbelts reduce serious injuries and deaths by about half (when used). Seat belt - Wikipedia. If the AP accident reduction numbers are anywhere close to what the NHTSA reported, it could mean that the first generation of Autopilot provides the greatest gains in driving safety since the seatbelt, and perhaps ever. And with the significantly improved hardware in AP2 and more fleet learning, the safety gains should grow from here.

It will be very interesting to see the data a year or two from now once AP1 and AP2 are further refined through fleet learning and software tweaks.

 

[EinSV]

Alex Roy wrote a great piece on the NHTSA's report, and its implications for the rest of the industry.

A couple highlights:

"NHTSA investigator Kareem Habib dismantles every argument critics and competitors have been firing at Tesla since Autopilot was released in October of 2015."

"Any hopes the legacy automakers might have had that regulators would throttle or halt Tesla’s progress are now shattered. What appeared to be Tesla’s headlong rush toward autonomy is now a three year head start."​

The NHTSA Report Exonerating Tesla Should Terrify the Auto Sector

 

[bhzmark]

Consumer Reports currently has this laughable absurd ALERT and criticism of Tesla's Autopilot in their review and rankings of Teslas:

2016 Tesla Model S | Reviews and Ratings from Consumer Reports

and

New & Used Car Ratings, Reviews & Buying Guides - Consumer Reports

It should say, "We previously thought that semi-autonomous driving packages were dangerous -- including Tesla's. But since we are an organization that purports to look at facts and evidence and we have serious real scientists in our employ, we published a retraction and welcome automakers that lead the way in implementing accident-reducing and life-saving driving assistance packages like Tesla's"

I'm a longtime subscriber to Consumer Reports -- and definitely will be cancelling unless they issue a retraction.

My Account - Online Sign In

 

[Spidy]

I don't see what's wrong with this. Should manufacturers make sure to market this as a assist system and not autonomous car and ensure it's also used in such a way? Of course they should. Should you keep your hands on the steering wheel? Well, that's exactly what Tesla tells you. Tesla even recently added a stronger warning and now disables your car if you don't put your hands on the wheel in time, sounds exactly what Consumer Reports was asking for and I doubt Tesla just did it because of them and rather did it because it made sense.

 

[bhzmark]

I don't see what's wrong with this. Should manufacturers make sure to market this as a assist system and not autonomous car and ensure it's also used in such a way? Of course they should. Should you keep your hands on the steering wheel? Well, that's exactly what Tesla tells you. Tesla even recently added a stronger warning and now disables your car if you don't put your hands on the wheel in time, sounds exactly what Consumer Reports was asking for and I doubt Tesla just did it because of them and rather did it because it made sense.

It's an "alert" just for the Tesla cars. As if something were wrong with them. And started from a longer CR article where Tesla was critical of Tesla.

Of course Tesla didn't do anything in response to CR nonsense.

 

[brucet999]

In its Autopilot report today, the NHTSA reported that accidents in AP-equipped Tesla's dropped by a remarkable 40% compared to Tesla's without AP.

According to the NHTSA data, AP-equipped Teslas appear to be almost twice as safe as Teslas w/o AP. And this likely understates the safety benefits of AP, since AP is only in use part of the time.

Let's not get too excited.

NHTSA did not say that Auto Pilot use resulted in 40% fewer accidents. They said that the accident rate decreased 40% since the wide introduction of Auto Pilot. The Auto Pilot capable cars were also the first with sensors for TACC and AEB. I submit that those capabilities have been most responsible for reducing the accident rate, not auto-steer; especially given the number of auto-steer-caused near misses (truck lust, off-ramp confusion, etc.) in the first several months.

This is a similar logical fallacy to Elon's comparison of accident rates per million miles running with Auto Pilot to the national average rate of accidents per million miles. There are more accidents per million miles on surface streets (mostly minor) and non-limited access highways than on the divided highways where AP is used, so of course the rate of accidents would be lower when AP is in use than the national average.

 

[brkaus]

I do know that with auto-steer engaged I found myself happy getting to my destination w/ the flow of traffic.

In my current car, without auto-steer, I frequently find myself switching lanes and driving a bit more aggressively... probably out of frustration. It gets me to my destination 2-3 minutes earlier. Gotta be more careful.

 

[Spidy]

It's an "alert" just for the Tesla cars. As if something were wrong with them. And started from a longer CR article where Tesla was critical of Tesla.

Of course Tesla didn't do anything in response to CR nonsense.

No it's not. E-Class has the same.

 

[brucet999]

Let's not get too excited.

NHTSA did not say that Auto Pilot use resulted in 40% fewer accidents. They said that the accident rate decreased 40% since the wide introduction of Auto Pilot. The Auto Pilot capable cars were also the first with sensors for TACC and AEB. I submit that those capabilities have been most responsible for reducing the accident rate, not auto-steer; especially given the number of auto-steer-caused near misses (truck lust, off-ramp confusion, etc.) in the first several months.

This is a similar logical fallacy to Elon's comparison of accident rates per million miles running with Auto Pilot to the national average rate of accidents per million miles. There are more accidents per million miles on surface streets (mostly minor) and non-limited access highways than on the divided highways where AP is used, so of course the rate of accidents would be lower when AP is in use than the national average.

Upon closer reading of the actual NHTSA report, I was wrong in the above. The report makes clear that they noted a 40% reduction of airbag events in cars with AP hardware after AP functionality was activated in comparison to the same cars before AP activation. The only difference between the data sets was AP having been activated or not.

 

[bhzmark]

Upon closer reading of the actual NHTSA report, I was wrong in the above. The report makes clear that they noted a 40% reduction of airbag events in cars with AP hardware after AP functionality was activated in comparison to the same cars before AP activation. The only difference between the data sets was AP having been activated or not.

That's right. And aeb and tacc were enabled before autosteer/autopilot was enabled.

E-class is harder because that autosteer is reported to be much worse and no data on whether it reduces accidents.

CR is simply wrong to suggest that Tesla's Autopilot cars are less safe or require some Alert to consumers.

 

[brucet999]

That's right. And aeb and tacc were enabled before autosteer/autopilot was enabled.

E-class is harder because that autosteer is reported to be much worse and no data on whether it reduces accidents.

CR is simply wrong to suggest that Tesla's Autopilot cars are less safe or require some Alert to consumers.

I wonder how big the dataset was that NHTSA used. We know roughly how many cars met the AP hardware/AP activated classifications, but how many airbag-deployment accidents were there among those cars; tens, hundreds, thousands?