hot take especially considering one has a class action lawsuit out and one does not
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Phantom braking so bad I want to return my car
- Thread starter RHINO SKIN SOL
- Start date
Well my sample size is TWO Subies if that makes a difference lol. Both vehicles have had zero issue. And I'm on the Subaru forums a lot. NO mentions of phantom braking there either. Compare and contrast that with what we see on the Tesla forums.“The lead plaintiffs, all current and former Subaru owners, say that EyeSight technology suffers from a number of dangerous vehicle safety defects. These defects allegedly cause the lane assist function in Subarus to engage without reason, as well as the automatic braking system. Their class action lawsuit contends that the carmaker knew of these defects, but still sold the cars, touting their safety and reliability. “
Nationwide Class Action Lawsuit Filed Over Subaru Eyesight Problems
Consumers say that Subaru vehicles equipped with EyeSight technology pose serious problems in a nationwide class action lawsuit.topclassactions.com
Oh Subaru system is so prefect, not.
Just like you told me if something works for you doesn’t mean it works for everyone, right?
I wish most of you stop repeating ignorant statements about others got it working and Tesla don’t. Which is not true at all. If you do more research everyone has some way or another glitches and problems in Driving Assistance systems. So Tesla is no different in that regard but at least thanks to FSD beta they are working on it and constantly improving the reliability of such system.
sleepydoc
Well-Known Member
Which is why I qualified my statement.
Here's how I came to make it:
We have had 0 AEB activations in roughly 10,000 miles of driving in the forester and probably 15 AEB activations in 15,000 miles on the model Y. If you calculate the number of activations per mile (or miles per activation) then yes, Tesla's system is orders of magnitude worse, even if you add margins of error.
It's possible that I got the one good Subaru and the one bad Tesla. From discussions here that doesn't seem to be the case but as I said, I don't feel I have enough information to truly make a conclusion.
I will point out that the NHTSA is investigating Tesla as well. Knowing how the legal system works, class action suits are relatively meaningless for anyone except the law firms that specialize in making money off of them. I'm more or less expecting one to be filed against Tesla, too.
Last edited:
Okay. Well Tesla does not. Not sure what else there is to say.
sleepydoc
Well-Known Member
)
charlesoris
2022 MYLR Red/White, 19", OD:12/13/21
Phantom braking is an extremely hard problem to tackle, and I wish Tesla only the best.
The weekend is just around the corner and I wish everyone love, peace, and happiness.
The weekend is just around the corner and I wish everyone love, peace, and happiness.
Hmmm, I seem to remember the guy with decades of radar experience saying that it isn't a radar problem causing PB in the vision + radar cars, it is an integration problem between radar and vision in the software... the exact same reason given by Tesla. Tesla gave up on a software solution for integrating vision and radar together and removed radar after using it for several years and making incremental improvements over time. Then they all of a sudden switch to vision only (in the "cheap cars", not the expansive ones) and say "we planned this all along, it has nothing to do with supply chain issues that every manufacturer is experiencing".
Oh, we all know that vision only is superior... that is why The S and X don't have it... you don't want the flagship cars to have the superior system right?
Keith
PS: Now, your job is to argue that they are experimenting with vision only on the "cheap" cars because they know the transition will have major problems while at the same time denying that these major problems that Tesla anticipated even exist... enjoy!
Last edited:
TheTheocracy
Member
Tesla's vision algorithm seems decently robust, though an improvement of 44.9% for VRU in the next beta means it still needs a good amount of improvement (only mitigates half of all phantom braking events). Most computer vision based applications I've seen and work with do have a good amount of false positives in an effort to have no false negatives, but they are usually random and can be mitigated using detections over time (temporal). We can 'kinda' assume that a false detection shouldn't occur in multiple frames.
Since multi-cam features (not temporal) are placed in the feature queue to update the spatial RNN (temporal), I wonder if false detections also update the spatial RNN, which leads to these phantom braking events. Almost like a lasting blip in the spatial location of the false detection, which requires the car to slow down. My guess is that they are 'training' that out using data (normal deep learning procedure) instead of denoising detections over time (more algorithm, more latency).
Since multi-cam features (not temporal) are placed in the feature queue to update the spatial RNN (temporal), I wonder if false detections also update the spatial RNN, which leads to these phantom braking events. Almost like a lasting blip in the spatial location of the false detection, which requires the car to slow down. My guess is that they are 'training' that out using data (normal deep learning procedure) instead of denoising detections over time (more algorithm, more latency).
PACEMD
Active Member
You knew what you were getting into when you purchased a Tesla, you are an unpaid Beta tester and you should like it that way..........Keith
buddhafree
2022 MYP
charlesoris
2022 MYLR Red/White, 19", OD:12/13/21
Your guy was right on the money. Data fusion of two dissimilar types of sensor systems (radar =active, visual = passive)is really tough, followed by data inference that requires enormous real-time on-board processing.
I am just going to add my personal observation from driving ~150 miles on 2 lane undivided highway in Idaho (that's what we have to go north and south).
We mostly used TACC (winding up and down canyons you don't want to do that). There were occasional slowdowns, maybe 2-5 mph, and I could tell 95% of the time what triggered them. Usually a larger vehicle when we were cresting a hill.
Adding Lane Keeping Assist (Beta) increased the slowdowns 20-fold. One or two major (from 65-50 mph) slowdowns. No slam on the brakes, terrifying stops at all. The same sort of thing seemed to trigger it, just way more often. Eventually I turned off Lane Keep (beta) and just used TACC.
Only once in that 150 miles was there a slowdown that I would call PB, and that was with Lane Keep. Again, not dangerous, just inconvenient.
Maybe that adds something to the discussion. Just my observation and I have a few more miles this trip.
We mostly used TACC (winding up and down canyons you don't want to do that). There were occasional slowdowns, maybe 2-5 mph, and I could tell 95% of the time what triggered them. Usually a larger vehicle when we were cresting a hill.
Adding Lane Keeping Assist (Beta) increased the slowdowns 20-fold. One or two major (from 65-50 mph) slowdowns. No slam on the brakes, terrifying stops at all. The same sort of thing seemed to trigger it, just way more often. Eventually I turned off Lane Keep (beta) and just used TACC.
Only once in that 150 miles was there a slowdown that I would call PB, and that was with Lane Keep. Again, not dangerous, just inconvenient.
Maybe that adds something to the discussion. Just my observation and I have a few more miles this trip.
That's an interesting observation. I have Lane Keeping Assist on. Maybe I should try turning it off.
Are we sure he meant that or he actually meant AutoSteer?
sleepydoc
Well-Known Member
Thanks - that roughly matches my experience in that the more automation you add, the more conservative (or scared) the car becomes. TACC < AP < FSD.
I'll also add that is suspect the 'scary slam on the brakes' episodes are false AEB activations. These are a problem, but a separate problem from phantom braking/slowing with TACC, AP or FSD. Most/all of the issues I've had with TACC have been ransom slowing of varying degrees of severity that is annoying but not dangerous.
I've said this before, but comparing experiences is very difficult with Teslas because software updates can make a big difference in how the car behaves. on top of that, some people have FSD beta and it's not clear how much overlap there is between the functionalities.
I assumed they meant AutoSteer. Lane Keep Assist is kind of 'AutoSteer lite.' Our Subaru has it and it will generally hold the lane but nothing else. It's also easier to overcome and deactivate than AutoSteer. If you turn on the turn signal it turns off completely while you change lanes and then reactivates once it's confident of the lane markings again. If you just turn the wheel it only takes mild force to overcome it compared to our Tesla where it feels like someone is actively resisting you with the steering wheel until it 'breaks free.' (note - I haven't used auto steer in any other makes, so they may be different)
(I have a hard time understanding the basis of the previously mentioned lawsuit against Subaru for auto steer spontaneously activating. If you're actually driving the car you'd hardly notice a difference and deactivating it is trivially easy, too, so any supposed safety issue would have to be highly exaggerated.)
Last edited:
Yes, Autosteer (beta) which is basically equivalent to lane keeping assist. Sorry for the confusion. Coming from Subaru I think of it as that.
I had another observation today (250 miles of dual lane). I had Autosteer enabled but not engaged and was using TACC. I still had those little slowdowns but not the big ones. Then, I tried disabling Autosteer (beta). Where you have to have the car in park to reenable it. I went the next 50 miles without even a slowdown. Then I had a couple, but nothing much. Very nice to drive. I have to think that the biggest cause of the slowdowns is Autosteer, not TACC. Even when it isn't engaged, but it is enabled, it seems to be doing something. Maybe chance, too but I don't think so.
When the slowdowns occurred I'd say 95% of the time a white (or silver) vehicle was coming at me in the other lane. Usually a truck. Once it was a big orange truck. The rest of the time big trucks, even red ones, it ignored.
We did have about 15 miles of divided highway driving for which we used TACC only, no Autosteer. No events at all.
So maybe Autosteer is baby FSB and that's where the majority of overcautiousness is coming from. Again, that's just our car but I would love it if someone else could try the same experiment.
stopcrazypp
Well-Known Member
Thanks for posting your tests. White trucks (especially semis) would match very well with the profile of that the trucks that have led to fatal accidents (or near fatal) with Teslas on AP. This adds more evidence to my point that Tesla may have tuned the system to be more "scared" of trucks.
Phantom Breaking so bad I want to return my car.
Phantom Breaking so bad I want to return my car.
Similar threads
