FUD I believe in. There is an enormous leap from a safe Level 2 system to a safe Level 3-5 system.

[Daniel in SD]

Since the technology all makes the car inherently safer, it all comes down to an argument that driver inattention when you switch from active safety features to full Autopilot is so awful, and the system so incapable of operating without constant driver attention, that it wipes out all these benefits.

Yep, that seems like the most probable reason if it does increase the number of accidents. That's exactly why Google decided against pursuing commercialization of their "AutoPilot" system.
So, what's the best way to go about determining if it's true or false?

 

[Daniel in SD]

Oh, I've read the paper.

The take-home message: "The central observations in the dataset is that drivers use Autopilot for 34.8% of their driven miles, and yet appear to maintain a relatively high degree of functional vigilance."

Do you believe that this chart from the paper accurately represents how people use Autopilot?

 

[DrDabbles]

Oh, I've read the paper.

The take-home message: "The central observations in the dataset is that drivers use Autopilot for 34.8% of their driven miles, and yet appear to maintain a relatively high degree of functional vigilance."

You should go watch Lex's videos. You might be very interested to see what his actual thoughts are, especially regarding the maturity and trustworthiness of autonomy, and the human reliance on systems they don't understand.

 

[EinSV]

Yep, that seems like the most probable reason if it does increase the number of accidents. That's exactly why Google decided against pursuing commercialization of their "AutoPilot" system.
So, what's the best way to go about determining if it's true or false?

Hmm, you skipped all the good stuff in my earlier post, so I'll recap:

Tesla cars equipped with industry leading active safety features built on Autopilot have 55% fewer accidents than Tesla cars without.

I was going to keep going but let's stop there for a minute.

Given that data, shouldn't all cars be required to be equipped with Autopilot for the active safety features alone?

Reducing accidents by 55% would save about 20,000 lives per year in the US.

Since you seem to be very concerned about safety I'm sure you'll agree with me that this would be a step in the right direction.

 

[Daniel in SD]

Given that data, shouldn't all cars be required to be equipped with Autopilot for the active safety features alone?

Huh? The active safety features have been standard on all Teslas for quite some time. That has nothing to do with Autopilot (TACC, Autosteer, and NoA). Active safety features improve safety. That’s why they’re standard on most new cars and will be legally required.
Can you explain the 55% number? How are you calculating that?

 

[EinSV]

Huh? The active safety features have been standard on all Teslas for quite some time. That has nothing to do with Autopilot (TACC, Autosteer, and NoA). Active safety features improve safety. That’s why they’re standard on most new cars and will be legally required.
Can you explain the 55% number? How are you calculating that?

Lol. Of course they have to do with Autopilot. AEB is part of Autopilot -- originally AEB+Autosteer was Autopilot, now AP is more complex and includes things like Obstacle Aware Acceleration (also part of the active safety assist procedures).

Do you think Tesla buys their active safety assist system from Bosch? It is built on Autopilot.

 

[Daniel in SD]

Lol. Of course they have to do with Autopilot. AEB is part of Autopilot -- originally AEB+Autosteer was Autopilot, now AP is more complex and includes things like Obstacle Aware Acceleration (also part of the active safety assist procedures).

Do you think Tesla buys their active safety assist system from Bosch? It is built on Autopilot.

They’ve always been standard whether you purchase Autopilot or not. Anyway it’s irrelevant since it’s quite clear I was never questioning whether or not they improved safety.
Also the context of this discussion is that someone brought up the Tesla safety report which makes a clear distinction between Autopilot and the active safety features. I feel that using Tesla’s terminology is appropriate for this discussion.

 

[AlanSubie4Life]

Oh, I've read the paper.

The take-home message: "The central observations in the dataset is that drivers use Autopilot for 34.8% of their driven miles, and yet appear to maintain a relatively high degree of functional vigilance."

I guess you have a HW1 vehicle (which most of the data is based on); that’s probably fairly safe, as it is perhaps less capable than HW3; as Fridman says, that should help you out, as it will keep you on guard. (Yes, I can see you actually have two Model 3s...)

I’m beginning to think from the responses to this thread that perhaps the good people of Tesla also really believe the data they publish shows that Autopilot is unassailably safer. It seems like it is quite “easy” to conclude Autopilot is safer with that presentation. Even pulled the wool over Elon’s eyes I guess!


To be clear (since it was in question I guess?): I agree active safety features are good! They all exist on all Tesla vehicles sold today, regardless of whether they are equipped with FSD.

 

[EinSV]

They’ve always been standard whether you purchase Autopilot or not. Anyway it’s irrelevant since it’s quite clear I was never questioning whether or not they improved safety.
Also the context of this discussion is that someone brought up the Tesla safety report which makes a clear distinction between Autopilot and the active safety features. I feel that using Tesla’s terminology is appropriate for this discussion.

Looks like verbal gymnastics to try to avoid saying something good about the Autopilot system -- that its active safety features are better than other automakers' and do a great job at avoiding accidents.

I think those are the most reasonable conclusions from the data and the NCAP studies.

I also think that has implications for the safety of the fully active Autopilot system as discussed in my earlier post.

I am far more concerned about the average driver on the road than I am with drivers behind the wheel of a Tesla with Autopilot, whether or not it is activated.

 

[EinSV]

I guess you have a HW1 vehicle (which most of the data is based on); that’s probably fairly safe, as it is perhaps less capable than HW3; as Fridman says, that should help you out, as it will keep you on guard. (Yes, I can see you actually have two Model 3s...)

I’m beginning to think from the responses to this thread that perhaps the good people of Tesla also really believe the data they publish shows that Autopilot is unassailably safer. It seems like it is quite “easy” to conclude Autopilot is safer with that presentation. Even pulled the wool over Elon’s eyes I guess!


To be clear (since it was in question I guess?): I agree active safety features are good! They all exist on all Tesla vehicles sold today, regardless of whether they are equipped with FSD.

I used to have an AP 1 Model S and we now have two HW 2.x Model 3s. I have Navigate on Autopilot set up to turn on automatically and use it 99% of the time I'm on the freeway.

I agree AP2 is better than AP1 but seriously doubt there is a significant difference in driver vigilance between the two systems.

 

[Daniel in SD]

Looks like verbal gymnastics to try to avoid saying something good about the Autopilot system -- that its active safety features are better than other automakers' and do a great job at avoiding accidents.

From Tesla's Q2 Safety Report:

In the 2nd quarter, we registered one accident for every 3.27 million miles driven in which drivers had Autopilot engaged. For those driving without Autopilot but with our active safety features, we registered one accident for every 2.19 million miles driven. For those driving without Autopilot and without our active safety features, we registered one accident for every 1.41 million miles driven. By comparison, NHTSA’s most recent data shows that in the United States there is an automobile crash every 498,000 miles.

 

[EinSV]

From Tesla's Q2 Safety Report:

Seriously, why is it so hard to acknowledge the obvious — that Tesla’s active safety features are enabled by Autopilot and part of the same system?


Active Safety Features
All Tesla cars made since October 2016 come standard with advanced active safety features for elevated protection at all times. These features are made possible by our Autopilot hardware system and include

Autopilot

And why is it so hard for Tesla critics to acknowledge what Euro NCAP found (and showed by releasing videos) - that the Autopilot-based active safety features are better than the competition?

Any discussion of Autopilot safety should acknowledge the strong evidence that the system reduces accidents even when in a more passive mode.

 

[Southpasfan]

Other than accidents which are by nature "unavoidable" that leaves all accidents which can be avoided, provided the driver is attentive. I would leave out of this category accidens which could only be avoided by professional stunt drive level of skill.

Human drivers are already inattentive. They sometimes drive while mildly to severely intoxicated. They eat and drink while driving. They look down to adjust music and air conditioning. They talk with passengers. They point at stuff out the window. On and on.

While its not a ridiculous question to ask whether the knowledge that the car you are driving has self driving features which are engaged could resultt in more inattentiveness, the answer is of two parts: (i) probably not, and (ii) if so, is the AP system in the car as attentive, or more attentive than the driver would have been in any event?

The incremental nature of the roll out of AP features means that the second part of the answer eventually completely encapsulates the first part!

Inattentivness, on its own, does not have a guaranted causal connection to accidents. Surely no one can argue about that.

All you have to do to see how this is going to go is to look at the progression between when AP required confirmation to change a lane, to the mid-2019 improvement where now it will change the lane unless the driver cancels the lane change.

For a human driver, not accurately checking the blind spot prior to a lane change is a classic bit of inattentiveness which results in accedents. Not all of the time, becuase sometimes the driver in the blind spot is attentive and avoids being hit.

It may not be simple to prove this "dog chasing the tail negative" of whether or not a self-driving system adds to overall inatentiveness, but Tesla was not, and will not be allowed to roll out an incremental feature like changing lanes without confirmation until the data shows that the car's sensors are at least as good, if not better, than humans are at checking the blind spot.

Once the sensors are as least as good (as in, NOW) it doesn't mattter if the feature makes drivers more inattentive! And by "doesn't matter" I mean does not matter in the sense of causing more accidents for a given opearational feature.

That's why the incremental nauture of the roll out is the key.

 

[Daniel in SD]

Seriously, why is it so hard to acknowledge the obvious — that Tesla’s active safety features are enabled by Autopilot and part of the same system?

I suggest you reread this thread because you appear to have lost track. This thread is not about active safety features, it's about systems like Autopilot as defined in Tesla's safety report. When discussing Tesla's safety report it is easiest if we use the terminology that Tesla uses.

And why is it so hard for Tesla critics to acknowledge what Euro NCAP found (and showed by releasing videos) - that the Autopilot-based active safety features are better than the competition?

I have acknowledged repeatedly in this thread that I believe the active safety features reduce accidents.

 

[EVNow]

FUD I believe in. There is an enormous leap from a safe Level 2 system to a safe Level 3-5 system.

LOL. Typical strawman.

L3/4/5 need error rates that are orders of magnitude better than the Tesla AP which is a L2 system.

Here's a video of what happened when Google gave their driver assist system to employees to test out. Note that their "AutoPilot" system is more advanced than what Tesla has released to the public. I still don't see how an advanced system that requires driver attention will ever be safe.

Waymo @ IAA Frankfurt 2019

How is this related to the title ?

 

[Daniel in SD]

How is this related to the title ?

The argument is that as Level 2 systems similar to Autopilot become more advanced they will become less safe due to human factors. There is a huge evolutionary leap required to get from a safe Level 2 system to a safe Level 3-5 system.
I admit my title is click-baity. I was just predicting the reactions from the usual suspects.

 

[EinSV]

Odd that can't acknowledge this simple fact:

Seriously, why is it so hard to acknowledge the obvious — that Tesla’s active safety features are enabled by Autopilot and part of the same system?

Or this one:

And why is it so hard for Tesla critics to acknowledge what Euro NCAP found (and showed by releasing videos) - that the Autopilot-based active safety features are better than the competition?

In any case, the point is simple:

(1) Since Autosteer maintains lanes far better than an average human driver it is not a big leap to believe that adding Autosteer to Tesla's best-in-the-industry active safety features further improves safety.
(2) Similar for Auto-lane change (even though it is not always subjectively "better" yet it appears to be as safe or safer than human lane changes).

As far as the subject of the thread, Levels 4 and 5 will only come with HW3, which we haven't really seen in action yet except in a limited way on Autonomy Day. Hard to judge until we do but based on the technical specs it should be a far more capable system.

If the idea is that even if the system is capable in the long run we won't get there because it is too dangerous to have people supervising an imperfect system while it is learning, I don't buy it. So far the theory that safety gains from improved Autopilot capabilities are automatically nullified by increased driver distraction is not really showing up in the data.

 

[Daniel in SD]

Odd that can't acknowledge this simple fact:

I feel like you're trying to derail this thread with things that are of no relevance. I have no idea what your point is. This thread is NOT about the active safety features! Start your own thread if you want to talk about them. I'll be happy to discuss them there.

So far the theory that safety gains from improved Autopilot capabilities are automatically nullified by increased driver distraction is not really showing up in the data.

I'm arguing that we don't have the data to determine that and you refuse to discuss the data that you have presented.

 

[AlanSubie4Life]

First, Autopilot is good. I am glad I have it, though probably Autosteer + TACC was really all I needed and could have saved some money. It’s not perfect by any stretch, but I anticipate making heavy use of it on road trips.

it is not a big leap to believe that adding Autosteer to Tesla's best-in-the-industry active safety features further improves safety.

Not a big leap, and literally everyone here (I think) concedes that this is possible. The point is there is just no clear data available which allows us to see whether it does - specifically there is no applicable comparison of data from non-Autopilot and Autopilot miles.

So far the theory that safety gains from improved Autopilot capabilities are automatically nullified by increased driver distraction is not really showing up in the data.

Would be good to show your work. Setting aside all confounding factors, Tesla’s own data shows using Autopilot reduces accident rates by 33% (2.19/3.27); can go 50% further between accidents. However, we know that Autopilot is used in different domains vs. where it is not. Are you sure that the baseline accident rate in those non-AP domains is no more than 33% higher than the Autopilot domains? They would need to be safer than that in order to demonstrate a potential AP benefit. I am certainly not sure! I don’t even know the exact domain differences! Maybe you have super-secret information I am not privy to?

Anyway, the point of the thread is that L2 Autopilot is getting better and better - and we are discussing the reasons why this trend to extremely high capability has caused other companies considerable concern. Not talking about active safety features - those are great and will be increasingly mandatory on all vehicles - and will get better and better with time.

 

[Southpasfan]

Tesla may release specific data sometime, but until FSD is fully developed I cannot imagine them doing so, as the data itself provides the key to implementation.

Further, I doubt that anyone but Tesla has the type of data some seem to be arguing is needed to make a definitive judgment on how safe AP is. To illustrate, I don't believe the NTSB has data concerning lane change decisions, and accidents caused by lane changes on a given stretch of freeway at a given level of traffic in similar weather and visibility conditions. To even write the sentence illustrates that such data does not exist.

Tesla, by comparison, does have enough cars out there, certainly in Southern California, for example, to actually have at least some data on lane change decisions in AP v. non-AP on the 210 Fwy between Pasadena and West Covina at rush hour. That data, that specific data, would allow comparison of the exact same type of lane change decisions so that one could compare the quality of the decision of human v. computer.

But we are not getting that. So the only way to look at this, I submit, is via the data there is. We know a couple of things. First, we know the number of Teslas out there gathering data, and we have accident statistics, which, in Tesla's case, can be broken down into AP v. non-AP driving miles. And those statistics show a gain in safety. To argue that accidents per mile isn't good enough is a fine argument, but since that is all we have I submit its a pointless argument.

But we have more. We also know that competitors, and investors taking short positions in Tesla stock, have every interest in the world, worth multiple millions of dollars, to point out any flaw they can find which would lead to a determination that AP was in fact dangerous.

The absence of the arguments which competitors and short sellers would make is evidence as well.

Plus, we also know how Tesla is rolling out features incrementally. Because of that, even us, as idiots on a message board, can speak to Tesla's actual operational AP performance, as long as we own one and drive it.

And this incremental roll out is genius or luck or both. Unlike Waymo, which is either going to have FSD or nothing, Tesla has a number of features that anyone who drives a Tesla can see clearly perform better than a human already. To get to FSD, you need to keep adding to those features, and I've seen additions in only a matter of months.