Phantom Braking

[Knightshade]

I haven't been able to find where you can't use Autosteer or TACC on a two lane highway.

The manual explicitly says it's intended for controlled access highways. This has been quoted in this very thread, several times.

My wife was nearly rear ended as a result of the phantom braking on 1 of 2 occasions.

Lots of folks think they were "nearly" rear ended.

Yet after 5+ years of AP2+ TACC being on the road, in a fleet of nearly 2 million cars, AFAIK zero people have been rear ended from phantom braking.

Almost like it's not nearly as dangerous as folks think or something.

 

[ohmman]

Yet after 5+ years of AP2+ TACC being on the road, in a fleet of nearly 2 million cars, AFAIK zero people have been rear ended from phantom braking.

Where have you looked? Or is this a case of "I haven't found something I've never looked for?"

I just did a quick search online and the first result was someone rear ended due to phantom braking. That doesn't mean it's endemic, but it does negate the "zero" you have been repeatedly stating.

 

[enemji]

Not doc, but I'll bite.

To me, it depends on the street/situation. If the street I'm on is residential and the speed limit is 25 mph, I will slow a tad as I go thru a green light to double check corners bcos you never know if a kid or dog may run out. If the street is 2-3 lanes in both directions and the speed limit is 45+, no way I'm slowing down.

Thanks for responding Big dog. I concur with you for the first situation but the second one also would slow down. Why? Because idiots who run the red lights are dime a dozen. I would take my foot of the gas pedal to make sure that I can brake faster in case there is the idiot running the red light

 

[Knightshade]

Where have you looked? Or is this a case of "I haven't found something I've never looked for?"

I just did a quick search online and the first result was someone rear ended due to phantom braking. That doesn't mean it's endemic, but it does negate the "zero" you have been repeatedly stating.

I was going by the years of reading threads on the topic here, and never seeing a single accident cited.

Also, to my knowledge, there's no accidents in the NHTSA reports either from this.... (and believe me, if there were, they'd have been mentioning in all the recent FUD stories about this)

In a quick search just now I did find: (note this wasn't near the top result, but it was halfway down the first page at least for me)

https://www.reddit.com/r/RealTesla/comments/iekn37

Which is one guy claiming it happened not to him but his dad.

But that's it.

So I guess I'll amend my statement to in all the years I've been reading about phantom braking, I've seen one third party claim of a single accident, with nobody injured.

 

[drtimhill]

Where have you looked? Or is this a case of "I haven't found something I've never looked for?"

I just did a quick search online and the first result was someone rear ended due to phantom braking. That doesn't mean it's endemic, but it does negate the "zero" you have been repeatedly stating.

But the point remains that for all the screaming here about how dangerous PB is, no-one here has reported any actual incidents of being rear ended as a result of PB (that' I'm aware of). And, people here are very vocal about Tesla flaws, so I'm sure it would have been reported.

Which makes you wonder how many times emergency braking has PREVENTED a Tesla rear-ending someone in an emergency?

Because that's the point. isnt it? Sure, ppl dont like PB, but how about when its REAL braking, not phantom? And please, no "well fix FB but keep real braking", the reality is any system that is trying to brake in a split-second emergency is by definition going to have some false positives.

 

[Xdama]

But as has been noted, Tesla are working on a fix .. its called FSD, which, as a side-effect of having to be much more sophisticated in understanding vehicles and oncoming trajectories, also handles phantom braking much more effectively. I strongly suspect their view is that this is the long-term solution (since even simple TACC will eventually use the same NN stack, even if you dont have the full FSD option).

What this means, imho, is that if enough people complain to NHTSA to force them to intervene, they wont get a quick fix at all .. they will get Tesla disabling TACC entirely until FSD is complete. So, "be careful what you ask for" guys.

Oh please disable it, turn it to dumb cruise only

 

[AndreP]

Yet after 5+ years of AP2+ TACC being on the road, in a fleet of nearly 2 million cars, AFAIK zero people have been rear ended from phantom braking.

Almost like it's not nearly as dangerous as folks think or something.

That's a little bit deceiving when the vast majority of Tesla vehicles have been delivered in the last 2-3 years, half of them in 2021 alone, and about 50,000 have no Autopilot hardware at all according to Lex Fridman's data:

Tesla Vehicle Deliveries and Autopilot Mileage Statistics - Lex Fridman

This page provides estimates of Tesla vehicles delivered and Autopilot miles driven segmented by Autopilot hardware version, based on the following milestones: Sep-2014: Autopilot Hardware 1.0 installed (Autopilot not enabled). Oct-2015: Autopilot enabled. Oct-2016: Autopilot Hardware 2.0...

lexfridman.com

And if we believe the reports to the NHTSA, it seems some stop using Autopilot entirely or use it very sparingly when phantom braking is bad enough to freak them out. So we might have some selection bias happening here.

There are so many factors that could be influencing what's happening.

 

[ohmman]

But the point remains that for all the screaming here about how dangerous PB is, no-one here has reported any actual incidents of being rear ended as a result of PB (that' I'm aware of). And, people here are very vocal about Tesla flaws, so I'm sure it would have been reported.

Which makes you wonder how many times emergency braking has PREVENTED a Tesla rear-ending someone in an emergency?

Because that's the point. isnt it? Sure, ppl dont like PB, but how about when its REAL braking, not phantom? And please, no "well fix FB but keep real braking", the reality is any system that is trying to brake in a split-second emergency is by definition going to have some false positives.

Are we conflating EAB and TACC here? Are they conflate-able? I understand them to be separate systems but don't know for sure. I did a search here and elsewhere, and the consensus appears to be that they're not the same system since you can use TACC with EAB disabled. So I don't know that we can argue for the EAB prevention as a hedge against TACC braking risk.

I concur that the data are slim on TACC related collisions, for Tesla and for other manufacturers. I found some papers discussing safety outcomes of TACC/Adaptive Cruise for all manufacturers and they have mixed conclusions - some saying it's worse, some saying it's better. Plus, they cover a lot of different manufacturers and implementations as well as stretch across a full decade. I could not find rear-end collision data by manufacturer or model, which may have provided a little insight. I'm sure it's available somewhere but I'm not sure I care enough to really hunt that down.

I don't agree that the lack of data proves the lack of events. Our group is a very, very small subsection of Tesla owners. In fact, I got a text from a friend while writing this about his brand new Plaid which shut down due to "complete electrical failure." I know others have reported that but he'll never be here to write about it, nor will he be on reddit or twitter or anywhere else. I also agree that the lack of data doesn't prove that there is a problem.

I think much of this just plays into my overall surprise that there aren't more accidents on the road in general. Have you ever considered all of the people you've met and wondered if, based on your interactions with them, you'd personally grant them a license to operate a deadly weapon on public roadways? Add to that the fact that a surprising percentage are under the influence of alcohol or other drugs and for me it's just shocking that we aren't seeing smashed up cars everywhere. I'm glad my intuition is wrong on this one..

 

[AndreP]

Lets add another caveat here that when it comes to NHTSA reports, only about ~1.1million Teslas have been sold in the United States specifically with the rest of them scattered across the world.

But most of the early Tesla deliveries were in the US, so that front-loads the US numbers a bit while also likely meaning that the 50,000 Teslas without Autopilot hardware are mostly or all in the US.

 

[Knightshade]

That's a little bit deceiving when the vast majority of Tesla vehicles have been delivered in the last 2-3 years, half of them in 2021 alone, and about 50,000 have no Autopilot hardware at all according to Lex Fridman's data:

Is it though?

I'm not sure "one 3rd party reddit accident out of a fleet of 1 million" cars is really massively different than "one 3rd party reddit accident report out of a fleet of 2 million"

Or even 0 reports out of a fleet of 500k (which would've been the situation in say 2019)

And if we believe the reports to the NHTSA, it seems some stop using Autopilot entirely or use it very sparingly when phantom braking is bad enough to freak them out. So we might have some selection bias happening here.

Given the NHTSA reports are a very very very tiny # I don't think even if they all quit using it it'd skew the #s noticably.

Tesla wouldn't be compiling the massive multi-billion # of miles of data they are if most people weren't using the system anymore.

Where's the accidents from this?



Contrast this with the Nissan example:

Nissan Rogue automatic emergency braking problems detailed - Autoblog

Some 800 drivers say their Nissan Rogue / Rogue Sport's automatic emergency braking system was triggered needlessly. 14 crashes are linked to the problem.

www.autoblog.com

Over 800 complaints and 14 NHTSA confirmed accidents, on a fleet of only ~550k vehicles, and over a period of just 2 years.


So even if we stuck to JUST the last 2 years for Tesla, the fleet BEGAN larger than that, is several times larger now, the number of complaints is much lower, and the number of NHTSA confirmed accidents appears to still be 0. Just one 3rd party reddit claim.

Lets add another caveat here that when it comes to NHTSA reports, only about ~1.1million Teslas have been sold in the United States specifically with the rest of them scattered across the world.

Safety regulators overseas tend to be MORE strict and heavyhanded than in the US.... (see how crippled the more advanced FSD features are there for example due to stricter safety regulation).

So that actually hurts the case for this being an actual danger.

 

[Big Dog]

Thanks for responding Big dog. I concur with you for the first situation but the second one also would slow down. Why? Because idiots who run the red lights are dime a dozen. I would take my foot of the gas pedal to make sure that I can brake faster in case there is the idiot running the red light

I see your point, but that still doesn't make sense to me. The vast majority of folks running red lights are punching it when they have a yellow, and it turns red right before (or after) they hit the intersection. But for you to get hit by them, you'd have to be starting in your lane just when your light turns green (as theirs just turned red). OTOH, if you are motoring down the Texas highway on a 2-3 lane road (each side) and the traffic is moving 50+ mph with cars ahead and behind you, and you can see a green light ahead, why would you slow?

 

[AndreP]

Is it though?

I'm not sure "one 3rd party reddit accident out of a fleet of 1 million" cars is really massively different than "one 3rd party reddit accident report out of a fleet of 2 million"

Or even 0 reports out of a fleet of 500k (which would've been the situation in say 2019)

Given the NHTSA reports are a very very very tiny # I don't think even if they all quit using it it'd skew the #s noticably.

Tesla wouldn't be compiling the massive multi-billion # of miles of data they are if most people weren't using the system anymore.

Where's the accidents from this?

Contrast this with the Nissan example:

Nissan Rogue automatic emergency braking problems detailed - Autoblog

Some 800 drivers say their Nissan Rogue / Rogue Sport's automatic emergency braking system was triggered needlessly. 14 crashes are linked to the problem.

www.autoblog.com

Over 800 complaints and 14 NHTSA confirmed accidents, on a fleet of only ~550k vehicles, and over a period of just 2 years.


So even if we stuck to JUST the last 2 years for Tesla, the fleet BEGAN larger than that, is several times larger now, the number of complaints is much lower, and the number of NHTSA confirmed accidents appears to still be 0. Just one 3rd party reddit claim.

Safety regulators overseas tend to be MORE strict and heavyhanded than in the US.... (see how crippled the more advanced FSD features are there for example due to stricter safety regulation).

So that actually hurts the case for this being an actual danger.

You seem to be focused on these Nissan stats like they're a rationalization for phantom braking in Teslas because there have been worse occurrences in the past, I'd view them as a cautionary tale and a reason the NHTSA needs to be proactive by building in solutions before 14 accidents occur.

Was phantom braking worse in general back in 2017-2018? Tesla barely had 550,000 vehicles on roads across the world by the very end of 2018, so the sheer number of Nissans driving when the technology was in a less polished state could be a contributing factor. Now we have the tech in 2022, and there seems to be a big uptick in reports across the Tesla models right now.

But there have not been 2million Tesla vehicles logging AP miles over the last five years and that's not how the regulators would look at it. They see a concerning trend as Tesla is massively ramping up deliveries, and concerning trends like this tend to precede worse outcomes. Based on the numbers of reports I'm seeing on Tesla vehicles posted to the NHTSA website in the last week alone, I'd expect those cited numbers to go much higher. We might see an update in the media next week as people watching this start counting up the February reports,

Unless you're digging into the websites for regulators around the world, we don't have visibility on what's being reported to them. But we'd likely hear about accidents through the media regardless of where they happened.

There are a lot of factors here dude and the nitty gritty is for the regulators to worry about. If people are experiencing issues that make them question the safety of the vehicle they're driving, they should be reporting them whether they're in a Tesla or a Nissan or a Chevy and the regulators can work with them to put effort into fixing the problems

 

[nvx1977]

I've had plenty of these events in my radar-equipped vehicles as well. I'll have a pure vision vehicle later this year and will see if it compares, but it's to the point where I am apprehensive about using it on the freeway. I nearly got rear-ended in the fast lane once and it really pissed off the person behind me.

I've also had it happen while using TACC and towing, which is a seriously freaky experience. My wife had an event on Interstate 80 in rural Nevada with nobody around while we had our Airstream, and it scared her enough that she will never use it with the camper again.

I'm not saying other manufacturers are better or worse, just saying that phantom braking is worth serious effort because it's scary and dangerous and it really undermines the huge benefit of TACC.

I shared your apprehension when my first-production Model 3 got FSD beta, which meant losing radar capability for regular AP. I have an 80-mile round trip commute, most of which is highway, and I've been driving this route since I got my Model 3 back in Feb 2018. In the past, I would have 1 or 2 phantom braking events per round trip due to signs and overpasses. I never saw any consistency with which signs or overpasses it would freak out at. They were all fair game. Sometimes the braking would be mild. Sometimes it was stronger than full regen.

When Tesla went vision-only, I saw posts here and articles stating definitively that phantom braking was a lot worse. So when I got FSD beta and went for that 80-mile commute, I was prepared for the worst. I kept my foot right above the accelerator the entire drive so that I could override any braking. Indeed, during the radar days, I got so good at reacting to phantom braking that I had my foot resting on the floormat. But with vision-only, I wasn't taking chances.

As it turned out, I had zero phantom braking for the entire round trip. I thought, maybe I got lucky. Eventually probability will even things out. But drive after drive, I had no phantom braking. After a few weeks, I stopped putting my foot over the accelerator pedal. It's been 3 months now that I've had FSD beta. I have had some minor braking events, but I wouldn't call them phantom. When a car in a neighboring lane drifts close to mine, sometimes my car slows, thinking that car is trying to merge into my lane. While I think the car was a bit too skittish, I don't consider this phantom braking. The car is just being overly safe. To me, these events are easily fixed, likely with single-stack v11.

I don't know why so many others are reporting a degradation in performance. For me, vision-only is an improvement from my radar days. There was a poll on here recently and people like me who experienced improvement with vision-only were in the minority, like 20%. Given we're all on the same firmware more or less, it has to be some sort of hardware variation at play. I hope your car is like mine

 

[particleman]

The manual explicitly says it's intended for controlled access highways. This has been quoted in this very thread, several times.





Lots of folks think they were "nearly" rear ended.

Yet after 5+ years of AP2+ TACC being on the road, in a fleet of nearly 2 million cars, AFAIK zero people have been rear ended from phantom braking.

Almost like it's not nearly as dangerous as folks think or something.

When the vehicle behind you has to leave the side of the roadway to miss you then you are damn near rear ended. That literally happened to my wife.

Again, TACC is fully supported on two lane highways. I've already pasted right from the manual, here it is again. We always used TACC, not autosteer.

"Traffic-Aware Cruise Control is primarily intended for driving on dry, straight roads, such as highways."

 

[Knightshade]

When the vehicle behind you has to leave the side of the roadway to miss you then you are damn near rear ended. That literally happened to my wife.

Not that it's any comfort to your wife of course, but anyone who has to leave the roadway to not rear-end a car braking in front of them was driving illegally in the first place.

Still, glad they were able to get out of the way and no accident occured.


This isn't the only place driving automation is going to make life interesting of course.... one well discussed elsewhere is that, legally, when an L3+ system comes out, it will have to obey all speed limits.

Which as others have remarked can actually be dangerous in areas where most folks drive like maniacs well over it.

Again, TACC is fully supported on two lane highways. I've already pasted right from the manual, here it is again. We always used TACC, not autosteer.

Ok.

But you specifically mentioned BOTH TACC and autosteer in the post to which I was replying there.

If you're now saying you never use autosteer (which is weird- it's awesome in the appropriate ODD)- why did you include it in the original question?



Here it is again if you don't recall doing that-

I haven't been able to find where you can't use Autosteer or TACC on a two lane highway. If that is the case Tesla should not allow it to be enabled.

Autosteer explicitly says to only use it on controlled access (ie divided and with on/off ramps) highways.

TACC does not.

They have different ODDs.

 

[_Redshift_]

It’s happened to me couple of times in the 3.5 years of owning my Model 3. Last time was on a freeway, no vehicles in the near vicinity, no overpasses, no shadows. I was doing 72, and slam! All of us lurched forward in our seats. Profanities flew at Tesla and Elon.

 

[Knightshade]

You seem to be focused on these Nissan stats like they're a rationalization for phantom braking in Teslas because there have been worse occurrences in the past

On the contrary- I'm using them to point out what a real safety defect looks like.

Far as we can tell reported accidents, from a fleet 1-4 times the size over the same period of time, has resulted in roughly 4-8x fewer complaints of even just unexpected braking at all, and infinity percent fewer accidents as NHTSA has had 0 reported... (and nobody has found any even anecdotal stories of actual accidents besides one 3rd party reddit post).

Was phantom braking worse in general back in 2017-2018? Tesla barely had 550,000 vehicles on roads across the world by the very end of 2018, so the sheer number of Nissans driving when the technology was in a less polished state could be a contributing factor.

Uh....it was the same # of nissans as Teslas at that time.


And yet Nissan had a bunch of accidents and 800+ NHTSA reports of problems.

Now we have the tech in 2022, and there seems to be a big uptick in reports across the Tesla models right now.

Again this is [B}grossly[/B] exaggerated by the November (just after the AEB phantom braking recall) number.

And even THEN the peak number was.... 51 complaints.

This dropped almost in half over the following 2 months.

Or to put it another way- in 2021 Tesla had 3-4x the fleet size Nissan did in 2019.

And yet even with this HUGE SURGE in reports, ended 2021 with roughly [B}8 times fewer complaints[/B} than Nissan had.

And again- 100% less actual accidents reported to NHTSA.

But there have not been 2million Tesla vehicles logging AP miles over the last five years and that's not how the regulators would look at it. They see a concerning trend

You keep claiming this.

And yet there's no open investigation by NHTSA on the issue. Even months after the one "surge" month, with complaints having dropped lower since then.

So it doesn't seem to be true.

So why keep claiming it?

as Tesla is massively ramping up deliveries, and concerning trends like this tend to precede worse outcomes.

Except the trend is down in the months following the big November pop after the recall.

And # of NHTSA reported accidents (ie actual outcomes in real life) is also still... 0.

Not very concerning it seems.


if the problem was significant, and was had been ongoing for a long while, and would get worse with fleet size, why wouldn't it keep going up?
(and why would it have "surged" coincidentally right after there was an ACTUAL defect that was already corrected)? What OTHER than the recall magically happened in October to cause a SURGE in reports?

Based on the numbers of reports I'm seeing on Tesla vehicles posted to the NHTSA website in the last week alone, I'd expect those cited numbers to go much higher. We might see an update in the media next week as people watching this start counting up the February reports,

This is an interesting point- because now that it's in the press, it's worth noting NHTSA doesn't verify the reports are from actual owners when filed.

And Tesla has been known to have false reports by non-owners filed to increase bad publicity.

So it might well be what's happening here.

In case you weren't aware this happens a lot-

The story behind the 'false' Tesla suspension complaints to NHTSA

Following reports of the National Highway Traffic Safety Administration (NHTSA) probing Tesla over a possible problem with the Model S’...

electrek.co

That's from 2016.

Where a single guy, who doesn't appear to own a Tesla at all, turned out to have filed 37 out of the 40 complaints against Tesla for a specific suspension safety issue. Which NHTSA -did- investigate. And found no actual safety issue.

Tesla report shows most negative NHTSA complaints are 'fake' and a tactic from 'detractors'

According to an email obtained by Electrek, Elon Musk shared with Tesla employees a report from last week that shows most NHTSA...

electrek.co

That's a story from 2018.

It mentions "After a review of hundreds of Tesla customer complaints filed with NHTSA since 2016 through present day in 2018, I can confirm at least two-thirds of them are fake"

Most of the fake complaints weren't even that hard to recognize, being posted by the same sources without valid VINs, often signed the same way...in some case cases with the author admitting they don't even own the car.




Again until there's any evidence of reported actually happened accidents I don't expect an sort of actual action/investigation by NHTSA.

Because right now all evidence is it's perhaps an annoyance, but not an actual safety defect.[/QUOTE]

 

[rxlawdude]

Lets add another caveat here that when it comes to NHTSA reports, only about ~1.1million Teslas have been sold in the United States specifically with the rest of them scattered across the world.

But most of the early Tesla deliveries were in the US, so that front-loads the US numbers a bit while also likely meaning that the 50,000 Teslas without Autopilot hardware are mostly or all in the US.

Except that would not explain how my July 2015 MS had a VIN in the 95,000 range. You think they sold around 45,000 MS between Sept '14 and July '15?

And yes, I recognize that VIN numbers are not necessarily in chronological sequence.

 

[Xdama]

Is it though?

I'm not sure "one 3rd party reddit accident out of a fleet of 1 million" cars is really massively different than "one 3rd party reddit accident report out of a fleet of 2 million"

Or even 0 reports out of a fleet of 500k (which would've been the situation in say 2019)




Given the NHTSA reports are a very very very tiny # I don't think even if they all quit using it it'd skew the #s noticably.

Tesla wouldn't be compiling the massive multi-billion # of miles of data they are if most people weren't using the system anymore.

Where's the accidents from this?



Contrast this with the Nissan example:

Nissan Rogue automatic emergency braking problems detailed - Autoblog

Some 800 drivers say their Nissan Rogue / Rogue Sport's automatic emergency braking system was triggered needlessly. 14 crashes are linked to the problem.

www.autoblog.com

Over 800 complaints and 14 NHTSA confirmed accidents, on a fleet of only ~550k vehicles, and over a period of just 2 years.


So even if we stuck to JUST the last 2 years for Tesla, the fleet BEGAN larger than that, is several times larger now, the number of complaints is much lower, and the number of NHTSA confirmed accidents appears to still be 0. Just one 3rd party reddit claim.






Safety regulators overseas tend to be MORE strict and heavyhanded than in the US.... (see how crippled the more advanced FSD features are there for example due to stricter safety regulation).

So that actually hurts the case for this being an actual danger.

As far as I know FSD isn’t allowed to be used in the UK, you can pay for it but not use it. I think it was due to be released last summer, not sure it happened though as there were queries as to who is responsible IF a fully autonomous vehicle causes an accident. The powers that be seem to think if in FSD a vehicles occupants are NOT resposible, it’s the car manufacture who is. Opens up a whole new ball game for car manufacturers with possibility of court cases etc. Someone else with a better understanding may clarify this though.

 

[BDC62]

As I posted earlier in the thread, on some rural highways here in Florida, I will have multiple PBs in a mile. On other rural highways, I will go multiple miles with zero PBs. I have stopped using TACC or FSD on the stretches of roads known to have PBs, and when I'm on roads with 1 or 2 PBs per mile, I disengage FSD whenever there is a vehicle following me closer than 1.5 to 2 seconds to avoid being rear-ended or inciting road rage of appearing to be constantly "checking my brakes". So, I will likely never be a statistic for being in an accident due to PB, but this does not mean I am having no issues with PB.