Does using AP increase safety?

[green1]

Ever since the wheel was invented we've been less safe.

Let's ignore all the other benefits and ban the wheel!

 

[S4WRXTTCS]

I think the final point you make is the key one IMO and I agree with you.

Whether you consider that technology "Autopilot" or part of the AEB is just semantics as far as I'm concerned.

Ha, it's the point I deleted since I didn't think it was really relevant.

What I was trying to get at is I see AEB/AP as being two distinct things. In my entire discussion on this thread about it I've considered them separate. AEB isn't what makes a Tesla different; AP is what makes a Tesla different.

AEB systems are being improved across the board on lots of cars from lots of different manufactures. The NHTSA is spearheading this at least in the US, and we'll soon start to see the tremendous difference it makes. It won't matter if it's a Subaru or a Tesla.

AP is a driving aid that will slowly morph into self-driving over various software and hardware upgrades. The usage of AP without accounting for AEB can enhance and diminish the safety of driving.

In this thread I listed the good and the bad of AP because the thread title said "Using AP". It didn't say AEB or any other always on collision mitigation/avoidance system.

Tesla themselves seem to be gathering lane steering data from people with AP engaged and without AP engaged. Where they're using this data to suggest the AP driving is better because it's more centered. The problem I have with that is I usually have a reason why I'm not centered, and it's not because I'm a bad driver.

 

[EinSV]

That's a faulty way to look at the data. It needs to be looked at under all circumstances in which AP was used because that's the way it WILL be used in reality.

As an extreme example, I could design a car with the brake and gas pedals reversed. Anytime someone got into an accident by hitting the wrong pedal, I could blame the driver for not using the vehicle as instructed and chalk it up to driver error. We know that's not the whole story though.

I love my AP and would be really upset if it were diminished but any discussions on the safety of AP needs to consider the Autopilot + human interface/behavior as a whole and not just look at the computer system.

Hota,

I agree with your main point (highlighted above). What I am trying to say is

• Year 1 of AP is done; and a new, improved version of AP is coming out
• Year 1 IMO was a huge success in terms of safety because
• The system was introduced into a complex environment and was generally safe
  • The limited data available suggests that drivers using AP are at least as safe as drivers not using the system.
  • There is no evidence of a single fatality caused by the use of the system as instructed

After Year 1, and with lots of experience and lots of data, Tesla is about to introduce a major upgrade. I am excited to see how it performs. There is every reason to expect that Elon's prediction that it will be 2-3X safer than driving without AP will turn out to be correct.

 

[EinSV]

Ha, it's the point I deleted since I didn't think it was really relevant.

What I was trying to get at is I see AEB/AP as being two distinct things. In my entire discussion on this thread about it I've considered them separate.

AEB systems are being improved across the board on lots of cars from lots of different manufactures. The NHTSA is spearheading this at least in the US, and we'll soon start to see the tremendous difference it makes. It won't matter if it's a Subaru or a Tesla.

AP is a driving aid that will slowly morph into self-driving over various software and hardware upgrades.

In this thread I listed the good and the bad of AP because the thread title said AP. It didn't say AEB or any other always on collision mitigation/avoidance system. It simply said AP, and to me AP needs to be engaged to count.

OK, this is getting a bit esoteric for me. I personally don't think it matters what is considered AEB and what is considered AP. I do think it is fair to consider what Tesla is contributing to safety/accident avoidance that the rest of the industry is not (in addition to convenience of using AP).

Elon says that he thinks that the new AP v.8.0 software will reduce accidents by 50% and improve safety by 2X-3X compared to vehicles that don't have it. Will be interesting to come back in 12-18 months and see what the data say about that. I expect the data will back up the predictions.

 

[S4WRXTTCS]

OK, this is getting a bit esoteric for me. I personally don't think it matters what is considered AEB and what is considered AP. I do think it is fair to consider what is Tesla contributing to safety/accident avoidance that the rest of the industry is not (in addition to convenience of using AP).

Elon says that he thinks that the new AP v.8.0 software will reduce accidents by 50% and safety by 2X-3X compared to vehicles that don't have it. Will be interesting to come back in 12-18 months and see what the data say about that.

There really have been huge strides in active safety within the automotive industry.

Even a lowly Subaru has an AEB system that can stop fully when the speed differential is less than 31mph. It has a stereo camera versus the single camera of the Tesla. There have been massive gains in safety throughout the industry due to the use of AEB systems.

AP exposed a massive limitation of the current AEB system in that it can't detect a stopped object all the time. It was exposed not because drivers were running into things on their own with AP not enabled. No one really paid much attention when that happen. It was exposed because AP was running into things while the driver wasn't paying attention, or simply watched in awe as their Tesla crashed into a stalled car.

Don't get me wrong because I do like what Tesla is doing. I think what they're doing is pretty unprecedented, and it's extremely exciting. They're able to improve the AEB through a software upgrade without having to upgrade the HW (like adding the tricameras).

I just don't like the statistical comparison that Elon Musk makes, and don't think to see an AP equipped system being 2X-3X better than a really good Volvo or MB.

As to thread I thought it was about using AP versus not using AP. As in are we safer to use AP? or are we more safe to leave it off where we instead use TACC only or don't use anything?

 

[Roger t]

Tesla AP logs 1.5 million miles a day. So it's likely there will be three or more accidents a year. It's inevitable. People who pay attention to the road will be safer with AP, and people who don't pay attention won't.

 

[green1]

and people who don't pay attention won't.

Even people who don't pay attention will be safer. Safer than they were when not paying attention without AP. They'd be even safer if they did pay attention, but they still benefit from the added safety of AP

 

[Roger t]

Even people who don't pay attention will be safer. Safer than they were when not paying attention without AP. They'd be even safer if they did pay attention, but they still benefit from the added safety of AP

Only if people's behavior stays the same. With AP, some drivers will pay even less attention to the road. Some may text longer, and some may drink and drive more. Many accidents are caused by those corner cases. So net gain in fatality may be less than we thought.

 

[3Victoria]

.... So net gain in fatality may be less than we thought.

But as long as it's a net gain, it's a win.

 

[Roger t]

But as long as it's a net gain, it's a win.

I think the biggest win of AP is to make driving much less stressful.

 

[S4WRXTTCS]

But as long as it's a net gain, it's a win.

To really have a net gain we have to improve ALL cars on the road in addition to the roads themselves.

Only a few of the fatality Tesla accidents could have been mitigated with the latest AP/AEB upgrade. it's a significant number, and a huge improvement. But, I think we'll start to see diminishing returns unless we improve ALL cars.

The saddest accident (at least in how I felt) was when an impaired driver hit a Tesla and the kid in the backseat of the Tesla died.

 

[Apricot]

I love Autopilot.

But since this is about data, this "AP logs 1.5 million miles" claim seems off to me. Tesla is selling about 14K cars per quarter. AP hit in the last quarter of 2014, so that's generously 14K x 8 quarters of AP equipped cars = 110K AP capable cars. Not all of them have AP on, so just guessing generously, say we have 100K AP cars out there. That would mean every AP capable car is averaging 15 miles per day ON AP ? That's a lot.

And, those calculations were pretty conservative above... TSLA has not been selling 14K cars per quarter since 2014, right?

 

[sillydriver]

There are four primary negative elements of the current version of AP that hampers it's safety. I'm only referring to the full AP, and not just TACC.

1.) Abuse in that drivers ignore how you're supposed to use AP. It works deceptively well for what little sensor technology is actually there. This deception leads some experienced drivers into assuming they know it's limitations/etc, but then they don't account for some once in a blue moon event. Something like trust lust, or a stalled car that suddenly appears in front of them.

2.) Using AP causes a lose of situational awareness. There is just something that happens when I allow the car to steer itself. I try to fight against that by at least having one hand/finger on the wheel for feedback. But, I still find my situational awareness to be lacking compared to what it is when I'm engage. This is precisely why lots of people feel relaxed and energized after a road trip with AP. They do because all that energy that went into situational awareness wasn't used. So of course you're going to feel relaxed.

3.) Control hand off - In the current AP there have been drivers assuming that AP was still active, but it's wasn't. That they disabled it or overrode it without realizing it. This is why the logs are so important when assessing why an incident happened.

4.) AP allows a person to drive considerably more miles without the same level of exhaustion, and this leads people to drive way more miles than they would otherwise. The chance of an accident goes up as you're on the road more. It prevents people from taking safer options like getting on a plane. I drive WAY MORE now days now that I have a Tesla than I ever did before, and AP played a significant role in that.

There are some benefits that enhance the safety.

1.) Diminished road rage - AP turns you into a softy. Someone is riding their brakes? Oh, well. Or if people are constantly speeding up and slowing down? So what?

2.) Less speeding - Just set the speed and forget about it. No looking down at the speedometer wondering why you're doing 100mph.

3.) A co-pilot for distracted moments - We all have distracted moments and you're lying if you say you're never distracted. Sometimes it's a text in a bad moment. Or sometimes it's some beautiful scenery. Or maybe it's a spider that's crawling on you.

4.) Ceasing lane drifters. Even people who are paying attention sometimes just drift.

FWIW, looking at what was written overnight I thought these points were very additive to my original post.

 

[EinSV]

There really have been huge strides in active safety within the automotive industry.

Even a lowly Subaru has an AEB system that can stop fully when the speed differential is less than 31mph. It has a stereo camera versus the single camera of the Tesla. There have been massive gains in safety throughout the industry due to the use of AEB systems.

AP exposed a massive limitation of the current AEB system in that it can't detect a stopped object all the time. It was exposed not because drivers were running into things on their own with AP not enabled. No one really paid much attention when that happen. It was exposed because AP was running into things while the driver wasn't paying attention, or simply watched in awe as their Tesla crashed into a stalled car.

Don't get me wrong because I do like what Tesla is doing. I think what they're doing is pretty unprecedented, and it's extremely exciting. They're able to improve the AEB through a software upgrade without having to upgrade the HW (like adding the tricameras).

I just don't like the statistical comparison that Elon Musk makes, and don't think to see an AP equipped system being 2X-3X better than a really good Volvo or MB.

As to thread I thought it was about using AP versus not using AP. As in are we safer to use AP? or are we more safe to leave it off where we instead use TACC only or don't use anything?

Nice post.

I agree that the relevant test is how Tesla's with AP performs relative to other vehicles.

In the recent press call regarding AP v. 8.0, Elon predicted that with the improvements in the new software, Model S and X would be twice as safe as other vehicles, and with fleet learning could soon be three times as safe. Elon Musk sees 3x potential increase in safety with Tesla’s new Autopilot update

So while it is interesting and important to understand how well Autopilot performed in its first year, that is just a beginning. By design, the system (and Tesla) are learning very quickly, and Tesla is about to implement major improvements in a system that based on the limited data available so far already is just as safe if not safer than driving other manufacturers' vehicles that don't have AP.

And if AP 8.0 performs as predicted and doubles safety immediately and triples it in the near future compared to other vehicles, that will be an extraordinary and even historic improvement in vehicle safety.

 

[Hota]

The proper comparison should really be: is driving really distracted using AP safer than driving moderately distracted in a non-AP vehicle.

I argue that for the majority of drivers - especially as the model 3 is rolled out - AP will lead to extreme distraction so we have to take that into account.

This is why Tesla is working so hard to make sure AP minimizes additional distractions.

 

[sillydriver]

So a theme here is that the situations for using AP that increase safety most compared with not using it include texting, eating, dealing with screaming kids, playing with the car's byzantine audio controls, and other distractions or stupid behavior.

I think this conclusion is correct, but I hope I am not alone in finding it amusingly ironic ...

The situations where AP increases safety the most are generally those where the driver is violating the manual's instructions for AP's use by not being "prepared to take immediate action."

Unfortunately, this means the manual cannot suggest that these are the situations where AP has the greatest benefit to safety, since it would be contradicting itself!

Of course people are generally smart enough to figure out by themselves that these are good times to use AP.

But then if they use AP when dealing with distractions -- just as they should -- but they drive over the crest of a hill and the thing swerves and puts them in the bushes, they better not post the incident on TMC because we will all (correctly) criticize them for not being prepared to take over.

Very ironic.

 

[RogerHScott]

I love Autopilot.

But since this is about data, this "AP logs 1.5 million miles" claim seems off to me. Tesla is selling about 14K cars per quarter. AP hit in the last quarter of 2014, so that's generously 14K x 8 quarters of AP equipped cars = 110K AP capable cars. Not all of them have AP on, so just guessing generously, say we have 100K AP cars out there. That would mean every AP capable car is averaging 15 miles per day ON AP ? That's a lot.

And, those calculations were pretty conservative above... TSLA has not been selling 14K cars per quarter since 2014, right?

It is probably the case that most long distance travel done in AP-capable cars is done with AP enabled, so that will skew things. Enough
to "fix" the numbers? Hard to say for sure.

 

[Jevi_S85D]

One undeniable fact is that AP reaction time is low and consistent.
6 milliseconds I believe I've read. Human drivers use more than 200 and often 1000 ms to react.

 

[SureValla]

This is a great thread, lots of well-reasoned arguments that have so far remained civil.

Disclosure: I do not have a Tesla, nor have I ever driven in any car that has any autonomous functions, be they AEB or higher.

We have 5 types of drivers:
a) Attentive Human - Driver breaking no laws with good reaction time
b) Inattentive Human - Driver eating/checking texts/talking on phone/screaming at kids in the back
c) Machine (AP v7 or 8) - Human driver gets in the back seat and watches as the car goes down the highway
d) Attentive Human + Machine (AP v7 or 8) - Driver from a) above using AP v 7 or 8, ready to take control at any time and also scanning the road
e) Inattentive Human + Machine (AP v7 or 8) - Driver from b) above using AP v 7 or 8, who has no situational awareness of where the car is and is not ready to take control in <x seconds.

We are trying to rank the 5 drivers above from safest to most dangerous. I think Elon would rank them the following

1) Attentive Human + Machine (AP v7 or 8)
2) Inattentive Human + Machine (AP v7 or 8)
3) Attentive Human
4) Inattentive Human
5) Machine (AP v7 or 8)

I think there is some healthy debate over whether 2 is safer than 3 (or possibly even 4) but that's not what I'd like to debate.

Instead, I'd like to ask how likely it is that we actually end up with driver a) from above. I'm theorizing that using AP v 7 or 8 leads to driver b) more times than driver a) i.e. a quick glance to read a text w/o AP will turn into actually responding to that text w/ AP. A quick glance at a nice view out the driver window will lead to a driver taking a perfectly framed photo of that view. A discussion about a board game will lead to the driver actually playing that board game with a fellow passenger.

This is not an easy answer because none of the statistics put out by Tesla to date answer it for the various reasons discussed above.

 

[sillydriver]

This is a great thread, lots of well-reasoned arguments that have so far remained civil.

Disclosure: I do not have a Tesla, nor have I ever driven in any car that has any autonomous functions, be they AEB or higher.

We have 5 types of drivers:
a) Attentive Human - Driver breaking no laws with good reaction time
b) Inattentive Human - Driver eating/checking texts/talking on phone/screaming at kids in the back
c) Machine (AP v7 or 8) - Human driver gets in the back seat and watches as the car goes down the highway
d) Attentive Human + Machine (AP v7 or 8) - Driver from a) above using AP v 7 or 8, ready to take control at any time and also scanning the road
e) Inattentive Human + Machine (AP v7 or 8) - Driver from b) above using AP v 7 or 8, who has no situational awareness of where the car is and is not ready to take control in <x seconds.

We are trying to rank the 5 drivers above from safest to most dangerous. I think Elon would rank them the following

1) Attentive Human + Machine (AP v7 or 8)
2) Inattentive Human + Machine (AP v7 or 8)
3) Attentive Human
4) Inattentive Human
5) Machine (AP v7 or 8)

I think there is some healthy debate over whether 2 is safer than 3 or vice versa but that's not what I'd like to debate.

Instead, I'd like to ask how likely it is that we actually end up with driver a) from above. I'm theorizing that using AP v 7 or 8 leads to driver b) more times than driver a) i.e. a quick glance to read a text w/o AP will turn into actually responding to that text w/ AP. A quick glance at a nice view out the driver window will lead to a driver taking a perfectly framed photo of that view. A discussion about a board game will lead to the driver actually playing that board game with a fellow passenger.

This is not an easy answer because none of the statistics put out by Tesla to date answer it for the various reasons discussed above.

I interpret you as asking whether turning on AP leads to the alternative to 3) being 2) rather than 1). In other words, "what is the corrupting influence of AP?" Good question. I think Elon's complaint/observation from the press conference to the effect that the "expert" users of AP are the ones who ignore warnings is consistent with the corrupting influence view. Frankly, I can't answer from personal experience because although I love playing with AP when appropriate, I don't take long trips in the Tesla and I live in a rural area where AP cannot be used on most of the roads. My guess is that there is some corruption but not too much. I doubt those who have the discipline to be really attentive to begin with are going to radically change their behavior to the negative, particularly since cases like running into the truck remind us of the potential consequences. Although actually, AP may become more corrupting if the radar innovations in 8.0 are demonstrated to make AP more reliable. In fact that is almost certainly the case.