Phantom braking so bad I want to return my car

[sleepydoc]

To an extent, yes, but the poster was saying (basically) “It’s easy to fix cause my Subaru doesnt do this” which, in your analogy, would be equal to the Subaru having comfy seat belts that snapped during a real accident.

Not really. If you're trying to argue that PB exists because the emergency braking systems are over sensitive then relaxing them would make Tesla's no worse than any other car on the road. Like I said, the emergency braking systems with TACC are intended to be a backup for driver error, not a replacement.

The other problem with that argument is that PB is so intermittent and often late that it would not reliably prevent an accident anyway. My MY will routinely brake right as I pass a truck. Well, braking 3 meters before I hit the truck will do nothing. Now you're left with either a TACC system that is deeply flawed or an EBS system that is deeply flawed. pick your poison.

Additionally, the frequency of the events leads to desensitization. If they are EBS events, the false positive rate is so high that the instinctive response is to simply press the accelerator because it's always a false alarm. In many ways, that's worse than having a system that is less sensitive.

 

[drtimhill]

Not really. If you're trying to argue that PB exists because the emergency braking systems are over sensitive then relaxing them would make Tesla's no worse than any other car on the road. Like I said, the emergency braking systems with TACC are intended to be a backup for driver error, not a replacement.

The other problem with that argument is that PB is so intermittent and often late that it would not reliably prevent an accident anyway. My MY will routinely brake right as I pass a truck. Well, braking 3 meters before I hit the truck will do nothing. Now you're left with either a TACC system that is deeply flawed or an EBS system that is deeply flawed. pick your poison.

Additionally, the frequency of the events leads to desensitization. If they are EBS events, the false positive rate is so high that the instinctive response is to simply press the accelerator because it's always a false alarm. In many ways, that's worse than having a system that is less sensitive.

But we're drifting .. my original point was not to defend FB per se, it was to point out that the claim that it could be "easily fixed" and as "proof" that "Subaru didn't do it" was an invalid argument, since the unstated assumption was that the Subaru/Tesla both provided equivalent true positives (which others have indicated elsewhere in the thread has been shown not to be). Whether Tesla are correct in tuning the emergency braking as they have is a distinct (and valid) discussion.

 

[Fourdoor]

I skimmed through the article but didn’t take the time to read all 35 pages it in detail, but there are some key points -
First, the article looked at false negatives. Phantom braking is a problem with false positives. As anyone who works with detection systems knows, the two are related (look up ROC curves.) Designing a system that has low false negatives is easy if you don’t care about false positives.

Second, these are adaptive cruise control systems, they are not designed to be autonomous; (adaptive cruise is considered Level 1 on the SAE autonomy scale) By definition they require driver monitoring and intervention. The cars ‘failed’ the tests because the tests evaluated situations the systems weren’t designed to detect.

This actually goes to illustrate my first point and something many people suspect - the problem with TACC is that it is over sensitive, reacting to what it doesn’t need to. I’ve had many occasions where it applies the brakes within a few meters of the oncoming vehicle. Yeah, it both cars are driving 50 MPH and it doesn’t try to stop until 5 meters before impact it really doesn’t matter. In the simulated test above where the cars were driving 15 and 25 mph. Even in this artifically low-speed and easier test, the Tesla failed to stop in time, so one could say that it failed, too. Of course it didn‘t, it would have required driver intervention, just like the other cars. But that’s the point. TACC is designed to require continuous driver engagement and driver intervention in emergencies.

I could also point out what some of the Tesla defenders tried to use as a defense earlier on - that the systems are only intended for highway use. If you use that criteria, then the oncoming traffic and bicycle tests wouldn‘t apply. Of course my Model Y will still slow down 15 MPH on a highway for no reason, but that‘s the entire problem here, isn’t it?

So yeah, this doesn’t really say that much, other than giving something for the fanboy trolls to regurgitate without actually thinking.

You are still having PB problems? Mine was resolved almost completely by the latest "oh crap, NHTS is looking at us for PB problems" updates. I am immensely pleased with the results and now love driving my car on my daily commute. I now probably experience the rare and minor PB that the PB deniers were experiencing all along

Anyone think that the improvements in eliminating PB were not due to the launched NHTSA PB investigation? I believe (perhaps incorrectly) that this was not a result of incremental changes from improved AI... I think they turned down the sensitivity by a large factor and will slowly crank it back up as they make improvements to the AI.

I wonder how well a Tesla with the latest software that almost completely eliminates PB would do in the referenced testing. I think it would do much worse than the Tesla used in the test... and since I pay attention when driving that is a trade off I am perfectly happy with to get ride of PB.

Keith

 

[Fourdoor]

Earlier today I had it slow from 50 down to 40 on a 4 lane divided county road with no one around me. A few days ago I had it slow down from 60 to 45 on the highway, similar conditions.

So yeah, still. Problem. I’m on 2022.4.5.21 (FSD beta) which several versions behind. I’ve seen reports that the newer versions are improved, so I’m hopeful but don’t have any experience to confirm.

Ahhhh, you are the test mule to resolve the problem without turning down the sensitivity of the system... we will gain better safety in our cars as yours learns to detect head on collisions without giving PB issues.

Thank you for your sacrifice

Keith

 

[TheLex]

Because you are assuming that the systems are equally capable of true braking in an emergency. Tesla could solve PB overnight by tuning the system to be less sensitive .. but at the expense of it not braking when a true emergency arises. Is this what you want? Would you be ok with explaining that to people who suffer from crashes and/or personal injury because of the de-tuning just to keep you comfortable?

Sure, I don't like PB either, but I'd rather put up with that and know that the car is ever vigilant for a true emergency. Go watch the videos of Tesla's avoiding or mitigating accidents thanks to AEB, then show me any of a Subaru doing the same.

Subaru's Eyesight has pre-collision emergency braking and by all accounts it works. So I don't know what you're referring to when you're implying Subaru's Eyesight system doesn't work. I've experienced Eyesight on my Subaru in action and it works. Contrast that with the myriad of reports of Tesla's hitting objects. It's funny how the fanboy Tesla community acts as if Tesla can do no wrong. All of these systems are driving aids. They're not a replacement for the driver actually driving. Furthermore if Tesla's have PB due to a large truck coming around the corner on a two-lane road, then that's a problem that many other manufacturer's vehicles just don't have.

 

[drtimhill]

Subaru's Eyesight has pre-collision emergency braking and by all accounts it works. So I don't know what you're referring to when you're implying Subaru's Eyesight system doesn't work. I've experienced Eyesight on my Subaru in action and it works. Contrast that with the myriad of reports of Tesla's hitting objects. It's funny how the fanboy Tesla community acts as if Tesla can do no wrong. All of these systems are driving aids. They're not a replacement for the driver actually driving. Furthermore if Tesla's have PB due to a large truck coming around the corner on a two-lane road, then that's a problem that many other manufacturer's vehicles just don't have.

I didn't say any such thing, for all I know the Subaru system may be excellent. I merely noted the poster had made an unspoken assumption that it worked well (for true positives) as the (hidden) basis of his argument, with no information provided to backup that assumption. And I get worried about "I've seen it in action" as proof of how good the Subaru system is and "myriad reports of Tesla's hitting objects" as how bad the Tesla system is. Anecdotal evidence is always shaky at best, littered as it is with confirmation bias, cascading accumulations etc. I've no idea which is better, and the various attempts at comparing these systems have suffered from poor methodology or just been overwhelmed by the complexity of the problem .. what speeds do you test at? What road conditions? What should the car do? What is a pass or fail? The Tesla system, for example, can attempt to mitigate accidents by steering the car. How can you compare that to systems that only apply the brakes? Which is "better"? What are the criteria for "better"?

 

[MP3Mike]

So I don't know what you're referring to when you're implying Subaru's Eyesight system doesn't work.

They are probably referring to the AAA study that shows that it doesn't work. (At least in the conditions they tested.)

 

[sleepydoc]

I also posted this in the AP/Phantom Braking thread, but it's relevant here I'll put it here, too.

One thing I've noticed recently is the car tends to brake when someone is approaching and passing you on the left. I was driving to our cabin a few weeks ago on a 2 lane state highway and on 2 or 3 occasions when a car was passing me the car would brake fairly aggressively about the time the passing car entered my blind spot. I noticed the same behavior again on the interstate the other day - I was in the middle lane and a car approached and passed in the left lane going about 10 MPH faster than me. In both cases the car was completely in the other lane, there was no traffic ahead of me or any other reason to brake. I will often let off the accelerator if someone is passing me on a 2 lane road to let them in, however Tesla actually started accelerating back to the set speed before the other car had finished passing me so I can't explain it by the car trying to be polite and that wouldn't apply to the interstate, anyway.

 

[WhiteWi]

I also posted this in the AP/Phantom Braking thread, but it's relevant here I'll put it here, too.

One thing I've noticed recently is the car tends to brake when someone is approaching and passing you on the left. I was driving to our cabin a few weeks ago on a 2 lane state highway and on 2 or 3 occasions when a car was passing me the car would brake fairly aggressively about the time the passing car entered my blind spot. I noticed the same behavior again on the interstate the other day - I was in the middle lane and a car approached and passed in the left lane going about 10 MPH faster than me. In both cases the car was completely in the other lane, there was no traffic ahead of me or any other reason to brake. I will often let off the accelerator if someone is passing me on a 2 lane road to let them in, however Tesla actually started accelerating back to the set speed before the other car had finished passing me so I can't explain it by the car trying to be polite and that wouldn't apply to the interstate, anyway.

In case you didn’t know you should alway slow down when someone passing you on two lane road on your left. Car is doing it what it‘s suppose to do.

 

[sleepydoc]

In case you didn’t know you should alway slow down when someone passing you on two lane road on your left. Car is doing it what it‘s suppose to do.

Why? I do it out of courtesy to allow the other driver to merge in sooner but I was never taught that in any driver's course and have never seen any laws stating such. (there've also been times I've had to slam on my brakes because the idiot passing me wast trying to do it when he didn't have room to do so.) Regardless, that doesn't seem to be why AP/TACC does it because it speeds up before the other car has completely passed me and it also seems to do it on the interstate.

It also seems to be related to how fast the car on my left is passing. When on the interstate and car is passing at a reasonable rate it doesn't seem to care. When the car is moving more quickly it seems to freak the Tesla out.

 

[WhiteWi]

Why? I do it out of courtesy to allow the other driver to merge in sooner but I was never taught that in any driver's course and have never seen any laws stating such. (there've also been times I've had to slam on my brakes because the idiot passing me wast trying to do it when he didn't have room to do so.) Regardless, that doesn't seem to be why AP/TACC does it because it speeds up before the other car has completely passed me and it also seems to do it on the interstate.

It also seems to be related to how fast the car on my left is passing. When on the interstate and car is passing at a reasonable rate it doesn't seem to care. When the car is moving more quickly it seems to freak the Tesla out.

Bc it’s the safe thing to do. Letting the driver pass you in opposite lane and come back to the right.

 

[TheLex]

They are probably referring to the AAA study that shows that it doesn't work. (At least in the conditions they tested.)

I realize it's a sample of one, but neither the Outback nor the Ascent have ever suffered from PB in my journeys all over the western US and Canada. And that includes LOTS of country roads.

 

[MP3Mike]

I realize it's a sample of one, but neither the Outback nor the Ascent have ever suffered from PB in my journeys all over the western US and Canada. And that includes LOTS of country roads.

That is probably because they tuned it to not have false positives, but the study shows that that results in false negatives. (That it doesn't react to imminent collisions.) More likely they designed it to only mitigate rear-ending the car in front, which is what I think regulations require, but they don't have it try to mitigate head-on collisions from oncoming traffic encroaching in your lane.

 

[sleepydoc]

That is probably because they tuned it to not have false positives, but the study shows that that results in false negatives. (That it doesn't react to imminent collisions.) More likely they designed it to only mitigate rear-ending the car in front, which is what I think regulations require, but they don't have it try to mitigate head-on collisions from oncoming traffic encroaching in your lane.

Yes. Adaptive cruise is not intended to be anything more than that; it's considered Level 1 automation. As such it requires full driver attention and engagement and accident avoidance isn't part or intended to be part of its feature set.

In reality, the tests above were a test of the sensitivity of the AEB systems. If you assume that the PB events are the result of Tesla's AEB system then it would seem Tesla has tuned it's AEB system to be more sensitive at the expense of more false positives affecting TACC. In reality, turning off AEB doesn't completely eliminate phantom braking so there there was more to it than that.

I realize it's a sample of one, but neither the Outback nor the Ascent have ever suffered from PB in my journeys all over the western US and Canada. And that includes LOTS of country roads.

Well, I can increase it to a sample of 4 because I and two of my coworkers have Subarus and have the same experience. In reality, though, you need to look at it not in terms of the number of cars but the failure rate per mile (or number of miles between failures.) Using this metric, Subaru's (and other makers') adaptive cruise is orders of magnitude better than Teslas.

Several people have reported that PB has improved with recent software updates. I'm on FSD beta meaning I'm behind but it looks like there should be an update coming out in the next couple of weeks so I'm eager to see if it improves.

 

[stopcrazypp]

I realize it's a sample of one, but neither the Outback nor the Ascent have ever suffered from PB in my journeys all over the western US and Canada. And that includes LOTS of country roads.

If you look at the study, where it failed is it failed to respond at all to threats coming from the opposing lane (while Tesla's system did). It's much easier to tune out false positives if you ignore the oncoming lane. Basically the two systems are trying to accomplish different things.

 

[sleepydoc]

Bc it’s the safe thing to do. Letting the driver pass you in opposite lane and come back to the right.

I think we’ve actually agreed on two things now!

 

[drtimhill]

I realize it's a sample of one, but neither the Outback nor the Ascent have ever suffered from PB in my journeys all over the western US and Canada. And that includes LOTS of country roads.

But that' the point here .. they dont do PB because (at least on the case of the Subaru), the referenced test showed it didn't brake AT ALL for some collision scenarios. So yes, it doesnt brake when it shouldn't, but it also doesnt brake when it should. So +10 points of comfort, -100 points for safety. Of course, this is only one test, and dozens (hundreds?) of different scenarios are possible in which the results could be very different.

 

[sleepydoc]

But that' the point here .. they dont do PB because (at least on the case of the Subaru), the referenced test showed it didn't brake AT ALL for some collision scenarios. So yes, it doesnt brake when it shouldn't, but it also doesnt brake when it should. So +10 points of comfort, -100 points for safety. Of course, this is only one test, and dozens (hundreds?) of different scenarios are possible in which the results could be very different.

It really depends on how you emphasize prioritize the different functionalities. On one hand you can say Tesla did a better job at detecting the potential collision and is therefore safer. On the other hand you can say the Subaru is designed for the human to be the primary safety system.

Then you can look at relative frequencies of events. An accident is worse than a phantom brake event; +10 points for comfort but -100 points for safety. Except if the accident event is rare and the Phantom brake event happens 1000x more frequently, how do you weigh them? Does it become +10,000 for comfort and -100 for safety? Or do you say the AEB is just a backup for the human, not the primary device, so even if it misses half of the events it’s still better than nothing. Ultimately that’s probably the big differences - Subaru is designing a backup system while Tesla is trying to designate primary system.

Of course, a Tesla isn’t treated the same as a Subaru if the AEB system fails, either. A Subaru having an accident because of an AEB miss might generate a NHTSA report whereas a Tesla stands a decent chance of being a headline. Part of that is Elon’s fault, but it’s still the reality Tesla has to deal with.

 

[WhiteWi]

I would rather Tesla save my life and life of my family in case of accident vs any amount of points for comfort but die in accident.

 

[TheLex]

But that' the point here .. they dont do PB because (at least on the case of the Subaru), the referenced test showed it didn't brake AT ALL for some collision scenarios. So yes, it doesnt brake when it shouldn't, but it also doesnt brake when it should. So +10 points of comfort, -100 points for safety. Of course, this is only one test, and dozens (hundreds?) of different scenarios are possible in which the results could be very different.

I'm not at all worried that my car isn't going to emergency brake for me in a severe accident. First of all, I am the primary safety device in such a scenario. I don't rely on the safety systems. However that being stated, as someone who spends a lot of time in SF Bay Area traffic where driving is a Roman Coliseum battle, I've had guys cut in front of me in heavy traffic and then hit their brakes, and my adaptive cruise braked faster than I could react. It's never failed me.

Also I've had occasions in parking lots where some kid literally ran in back of my car to chase a ball, and the Subaru stopped. That was impressive.

I'm not going to claim that some test found that Eyesight isn't fallible. It most likely isn't. But in my years of using Eyesight it hasn't failed me yet. Also if a semi truck with 80k lbs comes around a corner or crests a hill, even if your car stops completely, if that truck hits you, you're going to be in big trouble.