Time to trade ap1...elon says yes

[EinSV]

Not after this accident. The conclusion you stated was after the old crash where the car didn't see a truck crossing the road and the driver had a lot more time than 5 seconds to react. Since then Tesla implemented many more nags. This one shows the real danger, ignore your first nag and you can be dead. Maybe it will take another one where the AP smashed into the same kind of barrier without a nag just prior before people realize, hey, you have to watch it as if you were letting a toddler on your lap drive - which is more attention than just driving it yourself.

You are missing the point. The NHTSA concluded that after Tesla activated Autopilot serious (airbag inducing) accidents dropped 40 percent. According to Tesla, since that time their internal data shows Autopilot has become even more reliable. An Update on Last Week’s Accident

In the United States alone as of 2016 there were 102 fatal car accidents per day, or 37,461 per year, with fatal accidents on the rise in 2015 and 2016. Motor vehicle fatality rate in U.S. by year - Wikipedia

You can't conclude anything about a system's safety by looking at one accident in isolation while ignoring the fact that people get killed in automobile accidents in alarming numbers every day.

Claiming a system is unsafe when it reduces the number of serious accidents is absurd.

The details of the accident on Highway 101 you reference are being discussed in great detail in the thread you link: Model X Crash on US-101 (Mountain View, CA)

 

[lolachampcar]

Automated systems are already showing that they can reduce the number of accidents.
Automated systems' accidents will be different from human caused accidents and may, in some cases, have been prevented had a human been driving. Take for instance a human recognizing a pedestrian's focus and determining that they will step out into traffic. The human reacts to avoid the pedestrian where the machine can not ferit out that bit of information and hits the pedestrian. (imaginary example to make the point)

Until we collectively learn to asses and accept risk, people will not accept the machine killing someone when we would not have done so. This one is going to be a tough one.

 

[whitex]

You are missing the point. The NHTSA concluded that after Tesla activated Autopilot serious (airbag inducing) accidents dropped 40 percent. According to Tesla, since that time their internal data shows Autopilot has become even more reliable. An Update on Last Week’s Accident

In the United States alone as of 2016 there were 102 fatal car accidents per day, or 37,461 per year, with fatal accidents on the rise in 2015 and 2016. Motor vehicle fatality rate in U.S. by year - Wikipedia

You can't conclude anything about a system's safety by looking at one accident in isolation while ignoring the fact that people get killed in automobile accidents in alarming numbers every day.

Claiming a system is unsafe when it reduces the number of serious accidents is absurd.

The details of the accident on Highway 101 you reference are being discussed in great detail in the thread you link: Model X Crash on US-101 (Mountain View, CA)

Can you point me to where the NHTSA concluded the 40% drop in accidents? I know Elon did some back of the napkin calculations but those were torn apart by actual statisticians. I can't seem to find this one article where a statistician dumbs it down to show why those were meaning less, but one thing for example was any statistic where an extra handful of events (in this case an event would be a serious accident) changes the statistic completely, is meaningless. Another thing was that there wasn't really an apples to apples comparison, comparing Tesla vehicles with AP to the all vehicles on the road is not valid comparison. Bottom line, the statistic was a stretch spin by Elon (like my 691hp P85D). Anyways, if you have the NHTSA conclusion stating AP reduces serious accidents, please link it here.

 

[whitex]

Automated systems are already showing that they can reduce the number of accidents.
Automated systems' accidents will be different from human caused accidents and may, in some cases, have been prevented had a human been driving. Take for instance a human recognizing a pedestrian's focus and determining that they will step out into traffic. The human reacts to avoid the pedestrian where the machine can not ferit out that bit of information and hits the pedestrian. (imaginary example to make the point)

Until we collectively learn to asses and accept risk, people will not accept the machine killing someone when we would not have done so. This one is going to be a tough one.

While I agree with this rational, I do not believe it applies to Level 2 self driving (what EAP and AP is today), where a human must be prepared to take over in a split second. If the rate of intervention is every 1000 miles (and Tesla is nowhere near that), no way people will be ready, and most people will not want to play Russian roulette 10 times a year, even in exchange for no other accidents (which cannot be guaranteed either, but even if it did). I think the article Robot Cars Can’t Count on Us in an Emergency explains it well.

I do believe what you just said will apply to what Waymo and few others are doing. It will be interesting to watch society adoption to a autonomous transportation, for example, given a choice between Waymo's completely driver-less ride sharing in two years, and a human driven Uber or Lyft, if the price and availability is the same, which one will people choose? Is there a premium you can charge for driver-less, or will the price have to be significantly lower? I am watching in anticipation.

 

[EinSV]

Automated systems are already showing that they can reduce the number of accidents.
Automated systems' accidents will be different from human caused accidents and may, in some cases, have been prevented had a human been driving. Take for instance a human recognizing a pedestrian's focus and determining that they will step out into traffic. The human reacts to avoid the pedestrian where the machine can not ferit out that bit of information and hits the pedestrian. (imaginary example to make the point)

Until we collectively learn to asses and accept risk, people will not accept the machine killing someone when we would not have done so. This one is going to be a tough one.

So far regulators seem to be acting reasonably. They recognize the benefits of these systems in saving lives and seem willing to assess the benefits and risks fairly although there are never any guarantees in life.

Heavy handed regulation that would restrict customers' freedom to buy a feature that they want and that promotes safety is not only backward and irrational but downright un-American. Especially when a whole bunch of automakers -- not just Tesla -- are offering Level 2 systems or plan to soon.

 

[EinSV]

Can you point me to where the NHTSA concluded the 40% drop in accidents? I know Elon did some back of the napkin calculations but those were torn apart by actual statisticians. I can't seem to find this one article where a statistician dumbs it down to show why those were meaning less, but one thing for example was any statistic where an extra handful of events (in this case an event would be a serious accident) changes the statistic completely, is meaningless. Another thing was that there wasn't really an apples to apples comparison, comparing Tesla vehicles with AP to the all vehicles on the road is not valid comparison. Bottom line, the statistic was a stretch spin by Elon (like my 691hp P85D). Anyways, if you have the NHTSA conclusion stating AP reduces serious accidents, please link it here.

Sure -- a full copy was posted in this thread along with a discussion. NHTSA Finds Tesla Accident Rate Drops 40% After Autosteer Installed!

Here is a link to the NHTSA report itself:

NHTSA Full Report on Tesla Fatal Autopilot Crash | Tesla Model S | Road Traffic Safety

And here is the key figure from the NHTSA's report:

The NHTSA's discussion of the 40% reduction after AP was activated is in the attachment in the post quoted below.

This is not "spin" by Elon but a finding by the NHTSA.

Hope that helps. Since this is OT for this thread it may make sense to continue discussion in the NHTSA report thread linked above or one of the half a dozen or so threads discussing the recent accident but whatever suits you.

View attachment 211551

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.

 

[whitex]

Sure -- a full copy was posted in this thread along with a discussion. NHTSA Finds Tesla Accident Rate Drops 40% After Autosteer Installed!

Here is a link to the NHTSA report itself:

NHTSA Full Report on Tesla Fatal Autopilot Crash | Tesla Model S | Road Traffic Safety

And here is the key figure from the NHTSA's report:

View attachment 290897

The NHTSA's discussion of the 40% reduction after AP was activated is in the attachment in the post quoted below.

This is not "spin" by Elon but a finding by the NHTSA.

Hope that helps. Since this is OT for this thread it may make sense to continue discussion in the NHTSA report thread linked above or one of the half a dozen or so threads discussing the recent accident but whatever suits you.

Thanks for the reference. I've actually read it a while back. In that document NHTSA simply acknowledges that this was the data provided by Tesla. HOWEVER, AutoSteer was introduced together with FCW and AEB. According to NHTSA "FCW with AEB reduced rates of rear-end striking crash involvements with injuries by 42%" (source here). So, the 40% reduction for "FCW+EAB+AutoSteer" seems to be the same range as FCW+EAB alone (2% less actually).

Level 2 auto-pilot will continue to get more and more dangerous the better it becomes, as it will cause people to not pay attention.

 

[EinSV]

Thanks for the reference. I've actually read it a while back. In that document NHTSA simply acknowledges that this was the data provided by Tesla. HOWEVER, AutoSteer was introduced together with FCW and AEB. According to NHTSA "FCW with AEB reduced rates of rear-end striking crash involvements with injuries by 42%" (source here). So, the 40% reduction for "FCW+EAB+AutoSteer" seems to be the same range as FCW+EAB alone (2% less actually).

Nope, you are misreading the IIHS study you quote. The study actually concludes that FCW/AEB resulted in only a 2% decrease in overall injury accidents.

Take a second look at the portion I bolded above from the passage you cited -- the 42% reduction with AEB/FCW is only for one small subset of accidents (rear end collisions), and actually only half of those -- where the car with FCW/AEB rear-ended another car, not the other way around.

The 42% number excludes all other types of accidents, including head-on collisions, side swipes, single car crashes, and collisions in which the car with AEB/FCW was rear ended (including as a result of unnecessary or excessive braking caused by AEB or FCW)

In fact, the IIHS study found that there was only a 2% decrease in overall injury accidents with AEB plus forward collision warning and only a 6% decrease in overall accidents. Neither one even reached the level of statistical significance.. http://orfe.princeton.edu/~alaink/S...s/IIHS-CicchinoEffectivenessOfCWS-Jan2016.pdf (see pages 1 and 15 and also Table 3 on page 12)

The 2% reduction in injury accidents from FCW/AEB is paltry. Nothing compared to the 40% reduction in serious accidents the NHTSA found after AP was enabled.

It would be very helpful if you would acknowledge your error, because many people seem to make this same mistake and it causes a lot of confusion.

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

The only other automotive safety technology that has achieved safety gains of this magnitude is the seat belt. After 70 years of refinements since seat belts were introduced (1949), the CDC estimates that seatbelts reduce serious injuries and deaths by about half when used. Seat belt - Wikipedia.

If the NHTSA's numbers are confirmed, the first generation of Autopilot may have achieved roughly the same safety gains as 70 years of seat-belt technology. That is really remarkable, particularly given that the gains took place in a period when overall US traffic fatalities increased from 32,744 to 37,461. If those fatalities dropped 40% per vehicle mile instead of rising, there would have been only 20,938 fatalities in the U.S. in 2016. So 16,500 lives saved in one year alone.

While the NHTSA analysis is not conclusive (no study is) it certainly is promising and suggests there is very good reason to support the use of Autopilot (and similar systems) not just for convenience but to save lives.

Level 2 auto-pilot will continue to get more and more dangerous the better it becomes, as it will cause people to not pay attention.

I disagree. The data so far is very promising and with the significantly improved hardware in AP2 and more fleet learning, IMO safety gains are likely to continue to grow until the systems are reliable enough to become fully autonomous.

 

[The Duke]

Level 2 auto-pilot will continue to get more and more dangerous the better it becomes, as it will cause people to not pay attention.

The people who do not pay attention with auto steer are not paying attention without it. That's why AP is statistically safer. Those on their phones and even laptops are paying about the same attention whether or not they have have AP to assist.

 

[ricebucket]

If the NHTSA's numbers are confirmed, the first generation of Autopilot may have achieved roughly the same safety gains as 70 years of seat-belt technology. That is really remarkable, particularly given that the gains took place in a period when overall US traffic fatalities increased from 32,744 to 37,461. If those fatalities dropped 40% per vehicle mile instead of rising, there would have been only 20,938 fatalities in the U.S. in 2016. So 16,500 lives saved in one year alone.

While the NHTSA analysis is not conclusive (no study is) it certainly is promising and suggests there is very good reason to support the use of Autopilot (and similar systems) not just for convenience but to save lives.

I haven't read the study, but I'm skeptical. With these things, even small differences in methodology may bias things quite a bit and result in a misleading number.

The stat says 40% drop per vehicle mile. Specifically, off the top of my head, I have these questions:
1) Is this comparing all vehicle miles driven by vehicles without AP and vehicles with AP? Or is it comparing vehicle miles driven without AP turned on vs vehicle miles with AP turned on?
If the former, then the improvement may be due to other design aspects of Tesla vehicles and not AP. (For example, Teslas come with forward collision warning by default, but the feature is not standard on most other vehicles on the market.)

2) Does these number control for the types of roads driven? (This is more of an issue if the answer to the first question is the latter.) Autopilot may only be used on certain roads, which is a pretty selected subset of all roads. And the typical driver would be more likely to enable AP when road conditions are simple, resulting in an even more selected subset of roads. That is, miles driven with AP are more likely to occur on "easy" roads, where accidents are naturally unlikely. If the stat does not account for this disparity, then of course AP numbers will look much better.

3) You mentioned that this only considers accidents in which another vehicle is rear-ended. But what about other accident classes? The biggest news items on this forum have involved accidents with stationary objects (lane divider, fire truck). Hypothetically, it is possible that AP reduced rear-ending vehicles by 40%, but increased hitting stationary objects by a large number (e.g. 100%). If so, then the statistic would be highly misleading, as the accident rates for all accidents would look very different.

And these are just the 3 questions I came up with off the top of my head. I'm sure someone with more experience in these matters can find more questions in the methodology.

 

[mblakele]

I haven't read the study, but I'm skeptical

You can be as skeptical as you like, but you'll lose credibility if you don't do your own reading. Be an informed skeptic.

Here's a direct link to NHTSA, so you don't have to mess around with scribd. Look for "crash rates", p10.

https://static.nhtsa.gov/odi/inv/2016/INCLA-PE16007-7876.PDF

 

[ricebucket]

You can be as skeptical as you like, but you'll lose credibility if you don't do your own reading. Be an informed skeptic.

Here's a direct link to NHTSA, so you don't have to mess around with scribd. Look for "crash rates", p10.

https://static.nhtsa.gov/odi/inv/2016/INCLA-PE16007-7876.PDF

OK I didn't have time to read a dozen pages, but I did go and read the section you referenced. It does answer most of my questions, but I can still find more ways to poke holes at the methodology: The stats distinguish between cars before and after AP activation. Therefore the latter sample is cars that have been driven more on average. It could be that drivers have gotten more used to the car and its handling and so are thus less prone to accidents. For example, I had a few scratches on the side of my car within the first month of getting it because of how wide the car is. But since then I've not had any additional problems in that area, because I'm extra careful and aware of this fact.
Also, are both samples counting only cars with AP hardware (seems to be, which would be the correct thing to do, but at the expense of sample size)?

 

[whitex]

Nope, you are misreading the IIHS study you quote. The study actually concludes that FCW/AEB resulted in only a 2% decrease in overall injury accidents.

Take a second look at the portion I bolded above from the passage you cited -- the 42% reduction with AEB/FCW is only for one small subset of accidents (rear end collisions), and actually only half of those -- where the car with FCW/AEB rear-ended another car, not the other way around.

The 42% number excludes all other types of accidents, including head-on collisions, side swipes, single car crashes, and collisions in which the car with AEB/FCW was rear ended (including as a result of unnecessary or excessive braking caused by AEB or FCW)

In fact, the IIHS study found that there was only a 2% decrease in overall injury accidents with AEB plus forward collision warning and only a 6% decrease in overall accidents. Neither one even reached the level of statistical significance.. http://orfe.princeton.edu/~alaink/S...s/IIHS-CicchinoEffectivenessOfCWS-Jan2016.pdf (see pages 1 and 15 and also Table 3 on page 12)

The 2% reduction in injury accidents from FCW/AEB is paltry. Nothing compared to the 40% reduction in serious accidents the NHTSA found after AP was enabled.

It would be very helpful if you would acknowledge your error, because many people seem to make this same mistake and it causes a lot of confusion.

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

The only other automotive safety technology that has achieved safety gains of this magnitude is the seat belt. After 70 years of refinements since seat belts were introduced (1949), the CDC estimates that seatbelts reduce serious injuries and deaths by about half when used. Seat belt - Wikipedia.

If the NHTSA's numbers are confirmed, the first generation of Autopilot may have achieved roughly the same safety gains as 70 years of seat-belt technology. That is really remarkable, particularly given that the gains took place in a period when overall US traffic fatalities increased from 32,744 to 37,461. If those fatalities dropped 40% per vehicle mile instead of rising, there would have been only 20,938 fatalities in the U.S. in 2016. So 16,500 lives saved in one year alone.

While the NHTSA analysis is not conclusive (no study is) it certainly is promising and suggests there is very good reason to support the use of Autopilot (and similar systems) not just for convenience but to save lives.



I disagree. The data so far is very promising and with the significantly improved hardware in AP2 and more fleet learning, IMO safety gains are likely to continue to grow until the systems are reliable enough to become fully autonomous.

First, saying "If NHTSA's number are confirmed" and otherwise suggesting those are NHTSA numbers is misleading. Those are clearly labeled as "Tesla provided numbers" not independently collected statistics, say from insurance companies or DMV. Second, we really don't have the actual Tesla numbers to look at - how many cars, miles and accidents are we looking at here, statistics of few hundred accidents in a span of MY14-MY16 may not mean much. Third, again lacking apples to apples comparison, we don't know exactly what numbers Tesla considers serious accidents and how they compare with "serious" or "injury" accidents in the NHTSA study. Sorry, but given Tesla's history, I don't trust their interpretation of numbers - maybe what they meant to say is that AP is capable of 40% accident reduction, but currently limited by software, sensors or lack of compute power (this would an argument typical to Tesla, given their spin of numbers history with hp, battery capacities, etc.).

Any system that can kill you if you don't pay attention for a split second, and that same system at the same time is conducive to not paying attention, is plain dangerous, as Google engineers clearly concluded.

 

[Model S M.D.]

My thoughts exactly. I’m still enjoying my 12/15 build — the old nose, AP1, and 80 amp home charging 61 mph — see pic). I can’t get the latter with a new Tesla.
View attachment 288649

I figure that by the time they deliver on EAP, there will also be an internal design change based on lessons learned from model 3 and a bigger or more efficient battery. A year ago, didn’t Elon mention faster charging tech coming?

This.

I bet many of that 70% are waiting for AWD. I would not buy a car without AWD and many cars in this price segment are bought with AWD. Also some segment of those reservations are waiting for the $35K car. As a Tesla owner, you should be able to buy a Model 3 in a few months as they are releasing higher spec versions first.

Until all Model 3 variants are available, I don;t think we can really read into the demand equation too much.

When they offer AWD for the Model 3, Tesla will have to figure out a way to sweeten the Model S somehow with more updated tech or they will start losing Model S sales for model 3 sales. The flagship Model S is currently lacking some of the latest Tesla technology in the Model 3.

Things will get real interesting when AWD/Air suspension is released for the Model 3...

 

[Beryl]

But with even more improvements in the pipeline, sorry but you are just going to have to get over it.

I’m happy I bought my AP1 S in 12/2015 and enjoyed it’s smooth functionality all of this time but as a stockholder I’m cheering for AP2 improvements.

Level 2 auto-pilot will continue to get more and more dangerous the better it becomes, as it will cause people to not pay attention.

But there will be even fewer accidents overall.

(My dream car is the Model Y with FSD, air suspension, AWD & 72 amp charger).

 

[whitex]

But there will be even fewer accidents overall.

Yea, some day...

 

[mblakele]

First, saying "If NHTSA's number are confirmed" and otherwise suggesting those are NHTSA numbers is misleading. Those are clearly labeled as "Tesla provided numbers" not independently collected statistics, say from insurance companies or DMV. Second, we really don't have the actual Tesla numbers to look at - how many cars, miles and accidents are we looking at here, statistics of few hundred accidents in a span of MY14-MY16 may not mean much. Third, again lacking apples to apples comparison, we don't know exactly what numbers Tesla considers serious accidents and how they compare with "serious" or "injury" accidents in the NHTSA study. Sorry, but given Tesla's history, I don't trust their interpretation of numbers - maybe what they meant to say is that AP is capable of 40% accident reduction, but currently limited by software, sensors or lack of compute power (this would an argument typical to Tesla, given their spin of numbers history with hp, battery capacities, etc.).

Any system that can kill you if you don't pay attention for a split second, and that same system at the same time is conducive to not paying attention, is plain dangerous, as Google engineers clearly concluded.

Your criticism of the crash data section in the NHTSA report is reasonable. The numbers Tesla gave to the NHTSA may have a question mark next to them. We should qualify it with some sort of confidence measure.

But if you'll pardon me, I think you're going overboard in the other direction. Even if we throw out the "40% safer" claim completely, we can't jump to the conclusion that having AP is net more dangerous than not having AP. We don't have the data to support that conclusion. Instead we have a few isolated accidents, without any statistical basis for comparison: more anecdotes than data. Objectively, the most negative thing we can say is "needs more study".

Google / Waymo's decision not to ship an L2/L3 solution was different, and I don't think it's entirely relevant. They've targeted transportation as a service, with no safety driver. That demands L4 or L5, with a very high level of safety. If and when Tesla gets to that point, they'll also need to show a very high level of safety. But today, AutoPilot doesn't claim to be ready for that: it's still L2 driver assistance, and it may already have some safety benefits.

 

[EinSV]

First, saying "If NHTSA's number are confirmed" and otherwise suggesting those are NHTSA numbers is misleading. Those are clearly labeled as "Tesla provided numbers" not independently collected statistics, say from insurance companies or DMV. Second, we really don't have the actual Tesla numbers to look at - how many cars, miles and accidents are we looking at here, statistics of few hundred accidents in a span of MY14-MY16 may not mean much. Third, again lacking apples to apples comparison, we don't know exactly what numbers Tesla considers serious accidents and how they compare with "serious" or "injury" accidents in the NHTSA study. Sorry, but given Tesla's history, I don't trust their interpretation of numbers - maybe what they meant to say is that AP is capable of 40% accident reduction, but currently limited by software, sensors or lack of compute power (this would an argument typical to Tesla, given their spin of numbers history with hp, battery capacities, etc.).

OK, you do not mention it but it seems we are now in agreement that the 2% reduction in serious injury accidents from AEB/FCW are not in the same range as the 40% reduction in airbag-deploying(serious) accidents reported in the NHTSA study after AP was enabled.

I think that's where our agreement ends (assuming it exists there).

You say that the 40% reduction in serious (airbag deploying) accidents is "clearly labeled as 'Tesla provided numbers'" and not NHTSA's numbers. That's not correct. The NHTSA report makes clear that the NHTSA's Office of Defects Investigation (ODI), not Tesla, performed the analysis, using "Data logs, image files, and records related to the crashes" subpoenaed from Tesla:

"5.0 CRASH INCIDENTS 5.1 Autopilot crashes. ODI analyzed data from crashes of Tesla Model S and Model X vehicles involving airbag deployments that occurred while operating in, or within 15 seconds of transitioning from, Autopilot mode.14 Some crashes involved impacts from other vehicles striking the Tesla from various directions with little to no warning to the Tesla driver. Other crashes involved scenarios known to be outside of the state-of-technology for current-generation Level 1 or 2 systems, such as cut-ins, cut-outs and crossing path collisions."

"14 Data logs, image files, and records related to the crashes were provided by Tesla in response to NHTSA subpoenas.
​

https://static.nhtsa.gov/odi/inv/2016/INCLA-PE16007-7876.PDF (page 8)

The ODI, not Tesla, analyzed data of airbag deployments and crash rates:

5.4 Crash rates. ODI analyzed mileage and airbag deployment data supplied by Tesla for all MY 2014 through 2016 Model S and 2016 Model X vehicles equipped with the Autopilot Technology Package, either installed in the vehicle when sold or through an OTA update, to calculate crash rates by miles travelled prior to and after Autopilot installation. Figure 11 shows the rates calculated by ODI for airbag deployment crashes in the subject Tesla vehicles before and after Autosteer installation. The data show that the Tesla vehicles crash rate dropped by almost 40 percent after Autosteer installation.​

The ODI's analysis covered all Model Year 2014-2016 Model S and X with AP, a total of 43,781 vehicles according to page 1 of the report, and the ODI appears to have had access to a wealth of data. It used that data to investigate the exact concerns you have been raising, including "driver behavior factors" and "driver distraction" with AP (discussed on pages 9-10 of the report).

You say that you "don't trust Tesla's interpretation of numbers" but the report is the NHTSA's analysis of data logs, image files and other crash records provided by Tesla. It is the NHTSA's "interpretation," not Tesla's. The statement that "we don't know what Tesla considers serious accidents" is not relevant to the report -- the NHTSA did the analysis based on an objective criteria -- whether airbags deployed -- it did not rely on Tesla's assessment of what a serious accident would be.

So the bottom line is that the report makes very clear that the analysis, conclusions and numbers were all the NHTSA's based on data logs, image files and other records subpoenaed from Tesla.

As with any study there are limitations, as I said before. More data is always better -- maybe we'll get an update with more data along with the next NHTSA report. But a 40% reduction in serious (airbag deploying) accidents suggests the potential of this technology to save a very large number of lives, at a time when automotive fatality rates are rising in the U.S.

Ignoring the enormous life saving potential suggested by these results and claiming Autopilot is dangerous with no data to support the claim IMO is unjustified.

When I was growing up and seat belts were not yet mandatory some people were afraid to use them because they worried they might get trapped in a burning car. Not an irrational concern (like the concern about driver distraction with AP), but the numbers suggest that any risk is far outweighed by the benefits. Although the data is still limited, the initial results on Autopilot are very promising and suggest that any risks from increased driver distraction are likely to be far outweighed by the benefits.

 

[whitex]

Your criticism of the crash data section in the NHTSA report is reasonable. The numbers Tesla gave to the NHTSA may have a question mark next to them. We should qualify it with some sort of confidence measure.

But if you'll pardon me, I think you're going overboard in the other direction. Even if we throw out the "40% safer" claim completely, we can't jump to the conclusion that having AP is net more dangerous than not having AP. We don't have the data to support that conclusion. Instead we have a few isolated accidents, without any statistical basis for comparison: more anecdotes than data. Objectively, the most negative thing we can say is "needs more study".

Google / Waymo's decision not to ship an L2/L3 solution was different, and I don't think it's entirely relevant. They've targeted transportation as a service, with no safety driver. That demands L4 or L5, with a very high level of safety. If and when Tesla gets to that point, they'll also need to show a very high level of safety. But today, AutoPilot doesn't claim to be ready for that: it's still L2 driver assistance, and it may already have some safety benefits.

I agree with you that the available data is insufficient whether AP is more dangerous or safer. I was using Waymo's research as a guide where they concluded that "Level 2 cannot be safely implemented" rather than "statistically AP is more dangerous". As for Waymo's decision to target L4/L5, it's not quiet clear that it was their intention to do that from the start. It seems they wanted to do L2, then L3, but then decided to skip those at a later time, rather than planning to go straight to L4 from the very beginning. As a matter of fact, according to this article, they did a rapid shift away from L2 development in 2014.

 

[whitex]

Ignoring the enormous life saving potential suggested by these results and claiming Autopilot is dangerous with no data to support the claim IMO is unjustified.

Why did you not read the article I suggested earlier ( Robot Cars Can’t Count on Us in an Emergency ) It mentions for example data collected by Nauto, or Stanford research on how long it takes a human driver to regain control of the car, which is a the heart of the debate whether Level 2 can ever be safely implemented. Level 2 by the way requires the driver to regain control of the car at any time with little or even no warning at all (if AP is driving straight into a median divider, even if it doesn't warn you, you have to take over). This is the biggest problem with AP, it's the fact that it required the driver to babysit it at all times (and for those who don't, some get away with it, some pay for it).