Major Phantom Braking Issues

[stopcrazypp]

aah. do we have that officially or anecdotally ?

because "there might be something on the road - but I'll slam into it because someone is behind me" is a rather interesting algo for AP (...)

either there is something or there isn't ... AP taking you from 75mph to 50 mph within seconds clearly shows that the system thought "there is something"

Except with NN, there is a confidence level, it is not a binary decision. If no one is behind, the vehicle can safely err on the side of false positives. If someone is behind, then the confidence level of something in front must be much higher.

Part of this is also because if the car fails to respond to something in front, the car manufacturer is typically not liable. However if the car does an active maneuver that contributes to a accident (like slamming on the brakes unnecessarily), the car manufacturer may have partial liability if the car behind ends up crashing into you.

Anyways, I agree the above is just a theory, but as others point out, there's so many cases reported of PB happening when no one is behind, that it's fairly likely that is a factor (I previously pointed out it makes sense to use that as a factor the the decision, even when this pattern was not pointed out).

 

[enemji]

aah. do we have that officially or anecdotally ?

because "there might be something on the road - but I'll slam into it because someone is behind me" is a rather interesting algo for AP (...)

either there is something or there isn't ... AP taking you from 75mph to 50 mph within seconds clearly shows that the system thought "there is something"

Everything here is anecdotal. Even Musk will not confirm.

 

[enemji]

either there is something or there isn't ... AP taking you from 75mph to 50 mph within seconds clearly shows that the system thought "there is something"

Have you ever taken a GRE examination?

Have you ever seen the option on a multiple choice question?

“Not enough information to make a decision”

That is what is happening. Not what you are thinking.

The system does not have enough good data to make an informed decision to continue at the speed the vehicle is at.

 

[ItsNotAboutTheMoney]

Static cruise control is actually worse in every possible way, but we like it because we have adapted to using it, learning it quirks and cons and using it wisely.

I want static cruise control because:
- EV static cruise control is freaking awesome compared to ICEV due to precise and rapid control of motor RPM and strong regenerative braking. Neither of my plugins have adaptive cruise, and I live in a place with a lot of undulation, but I drive in cruise control almost all the time.
- Static cruise has no significant additional sensor dependency so it's always available.
- Static cruise does not suffer from phantom braking.
- We live in a relatively low-density location where the benefits of adaptive cruise control are limited: we just don't have people in front of us that often.

My wife wants static cruise control additionally because:
- People who can't maintain speed suck, and shouldn't be copied by driver assistance.

 

[enemji]

I want static cruise control because:
- EV static cruise control is freaking awesome compared to ICEV due to precise and rapid control of motor RPM and strong regenerative braking. Neither of my plugins have adaptive cruise, and I live in a place with a lot of undulation, but I drive in cruise control almost all the time.
- Static cruise has no significant additional sensor dependency so it's always available.
- Static cruise does not suffer from phantom braking.
- We live in a relatively low-density location where the benefits of adaptive cruise control are limited: we just don't have people in front of us that often.

My wife wants static cruise control additionally because:
- People who can't maintain speed suck, and shouldn't be copied by driver assistance.

I hear you but the priorities changed with the new gen (people and machines) and they started all over but this time focusing on high density city traffic so that one does not have to worry about stop and go traffic found in most metro areas.

I am sure they will get to addressing driving in low to no density roads like where you live.

 

[sjg98]

Well anecdotally, lots of people complain about PB and a good amount of those people also use a similar phrase to "glad there was no one behind me or they surely would have hit me".

And also, lots of people complain about PB but there doesn't seem to be lots of stories about accidents caused by it...you would think there would be lots of stories in the news if this was a problem that was resulting in crashes.

I am extremely doubtful of this assumption.
The assumed logic here is:
AP is dumb enough to hallucinate objects such that it PBs urgently.
AP is smart enough to judge distance of a following car reliably.
AP has logic such that it makes the decision not to brake when there is a following car?

What object could AP hallucinate that leads it to slow 25mph urgently if there is no following car, but doesn't brake really at all if there is a following car.
And if so - is that logic even remotely a good thing?

It's not like we have people reporting positive scenarios of "well it sorta-PBd, but much more gracefully because there was a following car".

 

[stopcrazypp]

I am extremely doubtful of this assumption.
The assumed logic here is:
AP is dumb enough to hallucinate objects such that it PBs urgently.
AP is smart enough to judge distance of a following car reliably.
AP has logic such that it makes the decision not to brake when there is a following car?

What object could AP hallucinate that leads it to slow 25mph urgently if there is no following car, but doesn't brake really at all if there is a following car.
And if so - is that logic even remotely a good thing?

It's not like we have people reporting positive scenarios of "well it sorta-PBd, but much more gracefully because there was a following car".

Judging if a car is following behind (regardless of distance) is a much easier task than trying calculate the probability of crashing in front.

We aren't theorizing it's some very advanced algorithm that accurately accounts that of the following distance of the vehicle behind, it could be as simple and dumb as:
If a car is not detected behind (regardless of distance or even perhaps lane position), brake/slow down even for objects in front even at lower confidence level.

 

[sandman1330]

Right. It is not stopping for no reason. It is just slowing down.
More importantly it will not PB when someone is following it.
So the situations are mutually exclusive.

Except that's not true. I've had it "brake check" people behind me several times in the last few weeks, one really bad one yesterday, and others have too (see cross posted link for background):

Has anyone heard of a Tesla being rear ended because of phantom braking?

Yes - she loves the car, but is quite sceptic on using the cruise For me too. TACC braked for a cyclist riding safely on side of road, braked for car parked safely off the road. At first I thought it was my job to anticipate these scenarios. But it was putting more stress on me in the form of...

teslamotorsclub.com

The reason no one has been rear ended yet is because even tailgaters often manage to avoid it, plus Tesla drivers generally manage to react with the accelerator quickly enough to avoid it.

And to the conversation about radar, I didn't have PB until they disabled my radar. Now I get it a lot.

 

[texas_star_TM3]

Except that's not true. I've had it "brake check" people behind me several times in the last few weeks, one really bad one yesterday, and others have too (see cross posted link for background):

Has anyone heard of a Tesla being rear ended because of phantom braking?

Yes - she loves the car, but is quite sceptic on using the cruise For me too. TACC braked for a cyclist riding safely on side of road, braked for car parked safely off the road. At first I thought it was my job to anticipate these scenarios. But it was putting more stress on me in the form of...

teslamotorsclub.com

The reason no one has been rear ended yet is because even tailgaters often manage to avoid it, plus Tesla drivers generally manage to react with the accelerator quickly enough to avoid it.

And to the conversation about radar, I didn't have PB until they disabled my radar. Now I get it a lot.

same here. would be much more useful if tesla vision would check with onboard radar (!) before doing a PB - versus (as some claim here) the defining factor for a PB event is if someone is behind you or not ... again - radar isn't fooled by hot air mirages over the asphalt

the amount of excuses being made here for "no radar needed - vision is all you need" is funny.

 

[KArnold]

If PB has a different "hit the brakes" threshold based upon whether there is a trailing car or not, you would think the "fix" would be easy - always use the higher threshold! But since we've had this ever since radar was dropped, it's apparently not an easy fix.

So perhaps the real problem is not "phantom braking" per se, but a bug in following distance that activates automatic braking with some level of TACC or higher engaged? Maybe Tesla needs more data on these relatively rare cases, hence the different collision threshold with no trailing traffic. *IF* true that's it's potentially more of a "collision" bug rather than a "braking" bug and *IF* it has different thresholds based upon a trailing car (very big *IF*'s), it implies that in some % of a very specific scenario, TACC+ will happily run into something it should not. Almost like the parked firetruck scenario which I believe has been long fixed.

Here's the rub that we'll never know. Let's say I'm on the freeway at 70 MPH, follow-distance two, on AP. I'm immediately behind someone else to the distance the automation allows. The car in front of me suddenly and aggressively changes one lane left due to a crashed vehicle in his lane noticed at the last moment. Surely we're not implying that the engagement of braking would be different depending on whether I had a trailing car behind me?? Of course not. Not to mention, I'm not even confident PB is related to AEB at all - you would think the AEB would be alarming as well if they were the same trigger.

Lots of speculation here, of course. I'm not sure having a different automatic braking action based upon whether there is a following car makes any sense to me.

So when will drop be to the first mere mortals begin for v11?

 

[sjg98]

Judging if a car is following behind (regardless of distance) is a much easier task than trying calculate the probability of crashing in front.

We aren't theorizing it's some very advanced algorithm that accurately accounts that of the following distance of the vehicle behind, it could be as simple and dumb as:
If a car is not detected behind (regardless of distance or even perhaps lane position), brake/slow down even for objects in front even at lower confidence level.

If the logic was this well controlled & simple, they'd have "fixed" it by now.

 

[stopcrazypp]

If the logic was this well controlled & simple, they'd have "fixed" it by now.

Note I'm not saying this is the main reason for PB and that disabling it would solve all PB, just speculating if such a function exists at all as a contributing factor.

There are a lot more variables that go into PB. The radar ones were usually overpasses, tunnels, and signs (which Tesla introduced a whitelist in the past to fix). There are also some unrelated to the sensors at all, but rather likely speed limit and curve databases (my currently radar based AP will lower the set speed at certain points in my trip, exactly at the same spots, and the common pattern is usually before a curve in an area where there are a couple of major highways converging).

 

[enemji]

Except that's not true. I've had it "brake check" people behind me several times in the last few weeks, one really bad one yesterday, and others have too (see cross posted link for background):

Has anyone heard of a Tesla being rear ended because of phantom braking?

Yes - she loves the car, but is quite sceptic on using the cruise For me too. TACC braked for a cyclist riding safely on side of road, braked for car parked safely off the road. At first I thought it was my job to anticipate these scenarios. But it was putting more stress on me in the form of...

teslamotorsclub.com

The reason no one has been rear ended yet is because even tailgaters often manage to avoid it, plus Tesla drivers generally manage to react with the accelerator quickly enough to avoid it.

And to the conversation about radar, I didn't have PB until they disabled my radar. Now I get it a lot.

What car is yours?

 

[enemji]

If PB has a different "hit the brakes" threshold based upon whether there is a trailing car or not, you would think the "fix" would be easy - always use the higher threshold! But since we've had this ever since radar was dropped, it's apparently not an easy fix.

So perhaps the real problem is not "phantom braking" per se, but a bug in following distance that activates automatic braking with some level of TACC or higher engaged? Maybe Tesla needs more data on these relatively rare cases, hence the different collision threshold with no trailing traffic. *IF* true that's it's potentially more of a "collision" bug rather than a "braking" bug and *IF* it has different thresholds based upon a trailing car (very big *IF*'s), it implies that in some % of a very specific scenario, TACC+ will happily run into something it should not. Almost like the parked firetruck scenario which I believe has been long fixed.

Here's the rub that we'll never know. Let's say I'm on the freeway at 70 MPH, follow-distance two, on AP. I'm immediately behind someone else to the distance the automation allows. The car in front of me suddenly and aggressively changes one lane left due to a crashed vehicle in his lane noticed at the last moment. Surely we're not implying that the engagement of braking would be different depending on whether I had a trailing car behind me?? Of course not. Not to mention, I'm not even confident PB is related to AEB at all - you would think the AEB would be alarming as well if they were the same trigger.

Lots of speculation here, of course. I'm not sure having a different automatic braking action based upon whether there is a following car makes any sense to me.

So when will drop be to the first mere mortals begin for v11?

*Evasive Maneuver* kicks into action I believe

 

[sandman1330]

What car is yours?

2022 Model S. Radar equipped but disabled.

 

[enemji]

2022 Model S. Radar equipped but disabled.

I have 2x2022 Model3 and none of them behave like what you described

 

[KArnold]

I have 2x2022 Model3 and none of them behave like what you described

It would be great for folks to post GPS coordinates (in a different post) for locations that are repeatable. I'd love to see a different car try the same if they happen to be nearby.

The only one I can repeat is the double stoplight slowdown Chuck has demonstrated.

 

[derotam]

Well anecdotally, lots of people complain about PB and a good amount of those people also use a similar phrase to "glad there was no one behind me or they surely would have hit me".

And also, lots of people complain about PB but there doesn't seem to be lots of stories about accidents caused by it...you would think there would be lots of stories in the news if this was a problem that was resulting in crashes.

I am extremely doubtful of this assumption.
The assumed logic here is:
AP is dumb enough to hallucinate objects such that it PBs urgently.
AP is smart enough to judge distance of a following car reliably.
AP has logic such that it makes the decision not to brake when there is a following car?

What object could AP hallucinate that leads it to slow 25mph urgently if there is no following car, but doesn't brake really at all if there is a following car.
And if so - is that logic even remotely a good thing?

It's not like we have people reporting positive scenarios of "well it sorta-PBd, but much more gracefully because there was a following car".

I am not assuming anything... I was purely responding to the question about whether there was anecdotal evidence

 

[derotam]

Except that's not true. I've had it "brake check" people behind me several times in the last few weeks, one really bad one yesterday, and others have too

Great, your issues seem like they may be REPEATABLE...care to provide dashcam video of some events? From the video we can calculate speed, distance, deceleration, following distance...etc and more fully analyze specific scenarios! "Really bad" can be a very subjective term.

Also if these issues are so repeatable for you, you could get an OBDII adapter and install ScanMyTesla app and record vehicle data during an event!

Ohh and I have said these same things to MULTIPLE people in the past and none of them came back with any additional video or other data.

The reason no one has been rear ended yet is because even tailgaters often manage to avoid it, plus Tesla drivers generally manage to react with the accelerator quickly enough to avoid it.

See now you are making excuses. People run into other cars every day because they are following too closely and/or not paying attention. With how many tesla's that are on the road all seemingly with a "MAJOR" PB issue that is soooo dangerous...one would think there would be crash data to support the reality of an actual danger...hence my previous comment about anecdotal evidence that seems to show that there aren't lots of rear end crashes due to PB.

It would be great for folks to post GPS coordinates (in a different post) for locations that are repeatable. I'd love to see a different car try the same if they happen to be nearby.

Exactly, as well as do what I said above and post dashcam video and/or provide SMT data to support.

I also love how people are trying to point to the other PB re-end collision thread but no where in that thread does anyone talk about a rear end crash....other than the SF Bay Bridge Tunnel....which is still under investigation.


Let me also make one thing clear... I believe PB is an issue that needs to be resolved and I think it does add a non zero risk for a rear end collision.

 

[sandman1330]

See now you are making excuses. People run into other cars every day because they are following too closely and/or not paying attention. With how many tesla's that are on the road all seemingly with a "MAJOR" PB issue that is soooo dangerous...one would think there would be crash data to support the reality of an actual danger...hence my previous comment about anecdotal evidence that seems to show that there aren't lots of rear end crashes due to PB.

Why are you approaching this with such hostility? I’m happy for you that you don’t experience this issue. Just because you have been lucky enough not to have it doesn’t mean it’s not an issue for others. Clearly it’s an issue for enough people to find itself discussed here repeatedly. In my case it happens on specific stretches when the conditions are just so, open highway ahead with the light reflecting off the road just the right way.

I didn’t happen to save the dashcam, and I have nothing to prove to you. Nor does anyone else, so the fact no one is jumping through your hoops to provide “proof” is irrelevant. We don’t need to prove anything to you.

Even if you assume I was not in danger of being rear ended (the fact that I intervened quickly prevented a rear ending, as is the case for most), it is still causing road rage issues. Again, I live in rural Alberta where the mere sight of a Tesla causes blood to boil for many. If they perceive I’m now brake checking them, the rage intensifies.

Like many others, I’ve had other cars with adaptive cruise (Golf R, Lexus). Whether you consider it a minor slow down or a major braking event, none of them do this, even minor slowdowns never happened out of turn. And all of them had the option to disable the ”adaptive” portion of the cruise and downgrade to regular cruise. That’s really all I’m asking for, the ability to disable the adaptive portion when those conditions exist that cause PB. As it stands right now I can’t even use cruise on certain roads in certain conditions.