Phantom Braking

[Knightshade]

Who else here thinks Tesla sells an inexpensive model?

So now you're gonna keep digging the hole deeper and suggest opinions trump actual numbers?

Almost 100% of Tesla vehicles produced in 2021 were the two cheapest options available regardless of trim

Why are you saying "regardless of trim"?


Is it because your original claim is factually wrong, and if you do not ignore trim that is obvious?


Dude.

You overgeneralized, incorrectly, and someone called you on it.

Be an adult, admit it, and move on.


Or, you know, keep digging I guess.... maybe tour the factory in Shanghai when you get the hole big enough.

All Tesla vehicles are expensive

Ah, keep digging it is!

No. Not all Teslas are expensive.

They sell a model below the average new car price right now

Your claim is factually wrong, and overly broad.


A story, in 3 parts:

 

[AndreP]

This threads incredibly toxic.

In its entirety, not even just focusing on the recent digressions

 

[Sporty]

“Had 4 events and a few more later on”. How in this sentence were you able to determine specifically 40 minutes? You might seem more credible without made up math data from assumption. Lol. A few of you are so committed to your mission of Being right you just start making up crap to feel important lol.

Forest. Trees. I think the point is that it doesn’t work well enough yet.

 

[sleepydoc]

I love how SURE you are of everything you say. There is no hint of doubt in your thoughts.

I clearly stated this is a nascent system that is evaluating more data than you chose to pay attention to. If only the programming was as simple based on 1 sensor and 1 parameter (distance).

And you you seem equally sure that there must be some legitimate reason that the system makes the car slow down. What evidence do you have? No one has given any evidence that these 'hidden dangers' actually exist, other than saying that "the system must be right."

On the other hand there are an abundance of examples where the system slows down for no reason in conditions where the drivers are able to clearly see and evaluate the risk. There are also systems made by virtually every other manufacturer that perform better. We also have statements from Tesla itself saying "phantom braking is a problem."

I have never claimed that the programming was simple, nor denied that Tesla's system was under development. I simply ask why we should put up with a poorly performing system. Yes, I am sure and no, I have no doubts that it can be done better. Given the evidence, how can you conclude anything else?

 

[OncomingStorm]

Comparing TACC to other Adaptive cruise is comparing two systems with identical functions. Comparing any adaptive cruise to 'dumb' cruise is absurd unless all you are trying to do is confuse the picture.

if expecting a feature to work at least as well as every other car on the road means I'm 'not for Tesla' then you're probably right.

I don't think you really understand how it works either. I don't say this as an insult - no one outside of Tesla does. you're hypothesizing about all the extraneous inputs confusing a complicated computer system and making it more prone to making mistakes but getting so caught up in the 'more complicated must be better' part that you seem to be oblivious to the 'making mistakes' part. Most significantly, you can't explain to me why any of it should matter.

I actually have an engineering background. I don't work for Tesla or in visual recognition, but I understand the issues. I also understand that the entire goal is to make a functional system. What I don't understand is you and others miss that point.

i just dont get your willful ignorance that tesla's system is vastly more complex than these cheap run of the mill systems that you keep groaning on about. more complex = more chances for errors (not an excuse, a fact), its that simple, no debate needed, i get that you are frustrated, and there defiantly are real problems we could be talking about, but you have to be willing to listen to simple reason or have an actual conversation about anything, so thats on me, i thought i could help shed some light that might help you understand and lessen your frustration but that clearly isn't working.

i do like how your argument is always "nope i'm right and your wrong here's the last 3 things i said" rofl, you can't just keep repeating the same garbage and expect anyone to listen. and i'm sorry you can have whatever background but read 2 of your posts and its clear you have little understanding of how software works as you keep talking like there is some switch they can flip to make you happy, its just not how it works. (not that you have to know anything, but its very clear your lack of understanding is contributing to your overall frustration)

anyway there is no point in continuing this so this will be my last post with you in mind, but i will say one final time.... if you truly are that unhappy with your car you really should sell it, these issues are not likely to magically go away any time soon and were I 10% as unhappy as you profess to be about mine i wouldn't sit with it any longer than i had to. i know everyone will take that as an attack but i truly think that, why own something that pisses you off? it makes no sence, especially right now, you could probably make a bunch of money on the transaction and buy another perfectly nice car without all the tech that makes you crazy. idk thats what i would do.

 

[Yelobird]

he said he used TACC for 60 minutes and had no issues for the first 20 minutes. 60-20 = 40. It's very simple math if you bother to read his post.

Actually no. He never defined the test time as 1 hour, you ASSumed that. He simply said the first 20 minutes of his drive were good. End of sentence. Then later had issues. Again, math only works with all the facts not assumed.

 

[sleepydoc]

Just took our new (non-fsd) model 3 on its first road trip. I had an hour of non congested road to try out AP.

It the first 20 minutes it worked great. I was very hopeful. Then I got 4 PBS then a couple more later on. The car slowed down from 70mph to 50 at distinctly more then normal regen but less then slamming on the brakes. Each time I checked the dash and the posted speed limit was correct. In the end I still mostly used it but there were 2 times where someone was following too close and I got nervous. I would reposition the car to a different spot. It’s useful but flawed. (As compared to any driving experience I have had in multiple other cars. All anecdotal to me. But I make many decisions based on my personal experiences).

Actually no. He never defined the test time as 1 hour, you ASSumed that. He simply said the first 20 minutes of his drive were good. End of sentence. Then later had issues. Again, math only works with all the facts not assumed.

See above.

 

[FirstInTown]

if you truly are that unhappy with your car you really should sell it, these issues are not likely to magically go away any time soon ......

I think someone told me the same thing a few pages back. Be careful what you wish for. Once I leave Tesla, I might not come back.

As it is, I could sell the MX for a gain and move over to driving the 2015 Caddy again, as there is nothing wrong with it. But... The world has a fever and we HAVE to stop buring our way to a hotter planet. As it is, Tesla is the only viable EV in our area. Sure I have a Volt and that does ok and is our other daily-driver for wife, but it is a car and Midwestern Americans 'need' our SUVs or trucks (kidding, but it seems that way). Chevy should have made the EquiVolt (Equinox Sized SUV with Voltec drivetrain), instead of Gen 2 Volt as a small car, but they didn't and killed it due to low sales.

There are 3 words that keep me from getting a Mach E, which my wife liked better than the Teslas when we test drove them a year ago. Super Charger Network! It kind of meets our needs, but a new Tesla SC planned for summer 2022 in Minoqua Wi will help going east in the cold. The rest of the EVs? Totally useless High Speed Charger Networks for Northern Wisconsin. EA long term, 5 year plan, showed ZERO new ones in 150 miles of us, so that sucks!

So I continue on with the X, driving with my foot instead of cruise when wife in the car because I am smooth and don't slam on the brakes for no reason. But. boy, would I like it if I had other EV choices that worked for us in Northern Wisconsin. Tesla has a 5-10 year lead? Maybe in that time other charge networks will come here, and I can look elsewhere. Looking forward to that day.

 

[sleepydoc]

i just dont get your willful ignorance that tesla's system is vastly more complex than these cheap run of the mill systems that you keep groaning on about. more complex = more chances for errors (not an excuse, a fact), its that simple, no debate needed, i get that you are frustrated, and there defiantly are real problems we could be talking about, but you have to be willing to listen to simple reason or have an actual conversation about anything, so thats on me, i thought i could help shed some light that might help you understand and lessen your frustration but that clearly isn't working.

i do like how your argument is always "nope i'm right and your wrong here's the last 3 things i said" rofl, you can't just keep repeating the same garbage and expect anyone to listen. and i'm sorry you can have whatever background but read 2 of your posts and its clear you have little understanding of how software works as you keep talking like there is some switch they can flip to make you happy, its just not how it works. (not that you have to know anything, but its very clear your lack of understanding is contributing to your overall frustration)

anyway there is no point in continuing this so this will be my last post with you in mind, but i will say one final time.... if you truly are that unhappy with your car you really should sell it, these issues are not likely to magically go away any time soon and were I 10% as unhappy as you profess to be about mine i wouldn't sit with it any longer than i had to. i know everyone will take that as an attack but i truly think that, why own something that pisses you off? it makes no sence, especially right now, you could probably make a bunch of money on the transaction and buy another perfectly nice car without all the tech that makes you crazy. idk thats what i would do.

Where do you get that I'm ignorant of Tesla's complexities? I've never said that, I just said that I don't care. You are right that more complex means more chances for failure, but again, what I care about are results. I keep repeating that because everyone else seems to keep ignoring it and making excuses.

I'm perfectly willing to listen, but as yet, no one has explained why Tesla's system is better. Please explain why it's better to have a complex system that performs worse at the same task than a simple system.

Are you saying that Tesla is taking more inputs than everyone else and will therefore have more processing errors because there's more bad data to weed out? If that's the case then they either need to have an algorithm capable of weeding out the garbage or reduce the input. Fundamentally, they should have the same minimal data set (or equivalent) that other cars have and so should be able to use that data just as effectively. After you have your basic functionality, the only reason to include more data is to either increase the accuracy or add more features. If it's the latter and the additional data is compromising accuracy then it should only be used for the additional features, not the primary one.

 

[sleepydoc]

I think someone told me the same thing a few pages back. Be careful what you wish for. Once I leave Tesla, I might not come back.

As it is, I could sell the MX for a gain and move over to driving the 2015 Caddy again, as there is nothing wrong with it. But... The world has a fever and we HAVE to stop buring our way to a hotter planet. As it is, Tesla is the only viable EV in our area. Sure I have a Volt and that does ok and is our other daily-driver for wife, but it is a car and Midwestern Americans 'need' our SUVs or trucks (kidding, but it seems that way). Chevy should have made the EquiVolt (Equinox Sized SUV with Voltec drivetrain), instead of Gen 2 Volt as a small car, but they didn't and killed it due to low sales.

There are 3 words that keep me from getting a Mach E, which my wife liked better than the Teslas when we test drove them a year ago. Super Charger Network! It kind of meets our needs, but a new Tesla SC planned for summer 2022 in Minoqua Wi will help going east in the cold. The rest of the EVs? Totally useless High Speed Charger Networks for Northern Wisconsin. EA long term, 5 year plan, showed ZERO new ones in 150 miles of us, so that sucks!

So I continue on with the X, driving with my foot instead of cruise when wife in the car because I am smooth and don't slam on the brakes for no reason. But. boy, would I like it if I had other EV choices that worked for us in Northern Wisconsin. Tesla has a 5-10 year lead? Maybe in that time other charge networks will come here, and I can look elsewhere. Looking forward to that day.

Yes. There are a lot of things I like about my MY and I fully expect things like autopilot and FSD to be buggy - they're in beta. I just can't accept that a feature that functions practically perfectly in other cars functions so poorly in my Tesla.

It was interesting - I was talking to some coworkers today and I mentioned how my MY's TACC worked worse than our Subaru's. A gal who just started last week said "that's interesting - we have the same car combination in our house and it's the same. The Subaru works fine but the Tesla keeps slowing down and slamming ion the brakes for no reason. It's really annoying." Hmm.

 

[Knightshade]

Where do you get that I'm ignorant of Tesla's complexities?

Well, you repeatedly insisting "the algorithm is broken" because the speed limit database, which has nothing to do with any algorithim, has bad data in it.

So that's a pretty clear clue you're ignorant of how the system works.

Plus every time someone tries to explain it you stomp your foot and yell how you don't care how or why it works you're just mad about it. So that's another clue.

I'm perfectly willing to listen

All evidence is to the contrary.

. Fundamentally, they should have the same minimal data set (or equivalent) that other cars have and so should be able to use that data just as effectively. After you have your basic functionality, the only reason to include more data is to either increase the accuracy or add more features. If it's the latter and the additional data is compromising accuracy then it should only be used for the additional features, not the primary one.

Yeah that entire paragraph belies a complete lack of understanding of...basically everything.

It's cool if you wish to remain ignorant of it. But you don't get to proudly parade that fact, then get mad you don't understand the multiple reasons people have given you for why it does things a certain way.

 

[Dewg]

Perhaps it would help if we renamed the technology from "Traffic Aware Cruise Control" to "Situationally Aware Cruise Control", since Tesla is taking into account more than just traffic. It also looks at map data, cameras read speed limit signs, etc. Other systems just look at the car in front of you and that's it. You set a speed, and it will hold that speed as best it can without crashing into the car in front of you. Tesla knows where all the cars are around you, even several lanes away. It knows your location with GPS, what speed the maps says you should have (if it can't find a speed limit sign), what speed you should be given a speed limit sign is visible, where you are heading via map data for navigation, and what lanes you should be in given your destination. It's a bit more than just trying not to hit the car ahead of you while holding the speed you told it to.

Before you jump down my throat, I'm not making excuses for Tesla's Phantom Braking problem, it definitely needs to be worked on and improved. I'm just trying to offer a broader view of Tesla's implementation versus similar technologies offered by other manufacturers.

 

[enemji]

And you you seem equally sure that there must be some legitimate reason that the system makes the car slow down. What evidence do you have? No one has given any evidence that these 'hidden dangers' actually exist, other than saying that "the system must be right."

On the other hand there are an abundance of examples where the system slows down for no reason in conditions where the drivers are able to clearly see and evaluate the risk. There are also systems made by virtually every other manufacturer that perform better. We also have statements from Tesla itself saying "phantom braking is a problem."

I have never claimed that the programming was simple, nor denied that Tesla's system was under development. I simply ask why we should put up with a poorly performing system. Yes, I am sure and no, I have no doubts that it can be done better. Given the evidence, how can you conclude anything else?

Thwnk you for the last paragraph. I agree it CAN be done better. That said, having spent a lifetime in industrial automation, I can say it is going to be a while., and also a roller coaster especially given the Tesla approach of agile releases.
Given the situation I might even be inclined to think that they are leaning on agressive safety actions rather than a smoother drive. Rather have a not so smooth drive than smoothly slam into a truck or a guardrail.

 

[danp]

FWIW, I just came back from a long weekend road trip in our vision-only Model 3 delivered in Dec, using non-FSD autopilot for about 5 hours total, exclusively on divided highways. It felt great, and not a single phantom braking event. In contrast, our older Model 3 with radar that the new 3 replaced got PBs on a regular basis on divided highways.

I've never tried autopilot on undivided roads. I'd find it more stressful, which is the exact opposite of what I want from driver assist features.

 

[S4WRXTTCS]

Perhaps it would help if we renamed the technology from "Traffic Aware Cruise Control" to "Situationally Aware Cruise Control", since Tesla is taking into account more than just traffic. It also looks at map data, cameras read speed limit signs, etc. Other systems just look at the car in front of you and that's it. You set a speed, and it will hold that speed as best it can without crashing into the car in front of you. Tesla knows where all the cars are around you, even several lanes away. It knows your location with GPS, what speed the maps says you should have (if it can't find a speed limit sign), what speed you should be given a speed limit sign is visible, where you are heading via map data for navigation, and what lanes you should be in given your destination. It's a bit more than just trying not to hit the car ahead of you while holding the speed you told it to.

Before you jump down my throat, I'm not making excuses for Tesla's Phantom Braking problem, it definitely needs to be worked on and improved. I'm just trying to offer a broader view of Tesla's implementation versus similar technologies offered by other manufacturers.

I think its important to clarify a couple of things regarding what's being used for TACC, and someone can clarify if I'm mistaken about any of it.

1. Map Data -> Uses the data to determine where to slow down for curves. Does this regardless of Vision+Radar or Tesla Vision.
2. Responding to Camera Read Speed Limit changes or map based speed limit changes -> Tesla Vision Only (with Vision+Radar I believe this is AP only)
3. GPS -> Used only for Map Data
4. Lanes you should be in -> Not used as its not needed for TACC
5. Cars in adjacent lanes -> Used to determine if there is a chance they'll cut into your lane, and used regardless of Vision+Radar or Tesla Vision.

All of the above ignores FSD Beta because its an entirely different stack

So I'm purposely limiting it to just TACC on a divided highway. This is something I have thousands of miles of experience with using Vision+Radar and thousands of miles with Tesla Vision.

In the early days of TACC a lot of the PB was caused by #5 with semi's, and I made some posts about really wanting a dumb Adaptive Cruise Control. Over time they improved it. At some point I started to use AP a lot more and most of my PB was related to #1 or #2. Sometimes I'd see the dreaded 45mph limit when passing though Tacoma. During the entire time with Vision+Radar I had a few occurrences of PB due to overhead bridges while there was no traffic in front of me.

So even before the Beta the PB was bad, but not horrible. Then I got the FSD Beta which brought Tesla Vision. I did hear all things people said about it, and I did see the PB poll on it but nothing prepared me for how bad it really was. Being in the Seattle area means a lot of rain, and it did worse in the rain which was really disappointing.

Thankfully the freeway performance did improve from 10.3.1 to 10.9, and PB might be down to the levels of Vision+Radar. I say might because there is a lot of randomness.

What I know is its extremely unlikely that ANYONE with Tesla Vision will be able to do a 400 mile trip from I5 near Everett all the way down to Portland, and back without at least a 3-4 PB events where one will be startling enough to take over to prevent further slowing. This is assuming the person drives 10mph over the limit, and has normal following settings, normal FCW setting, etc.

Would a Tesla Vehicle with Radar + Vision be able to do with? Possibly. I figure it has a better chance than 5%.

Can a 2015 AP1 do it where it has a lot of the same capabilities? Probably. I had one and PB was pretty rare despite also being able to adjust the speed for corners. I did feel like it was a little slow on its cornering speeds.

Can my Jeep Wrangler Unlimited do it where it doesn't have #1, #2, and probably limited #5 do it? Extremely likely.

Could I buy another vehicle that has 1, 2, 3, and 5? while having a good chance of no PB? It's a strong possibility. Here is the first google search I did, but there are probably a lot more.

This SUV can read speed limit signs and adjust accordingly

A pair of intriguing new features in the 2020 Lincoln Aviator luxury SUV promise to take some of the stress out of long drives.The Aviator is the first vehicle with:— […]

www.boston.com

 

[Yelobird]

See above.

And again, Where in his paragraph did he note 60 Minutes! That was made up in your head like most of the facts you present.

 

[Sporty]

And again, Where in his paragraph did he note 60 Minutes! That was made up in your head like most of the facts you present.

Just took our new (non-fsd) model 3 on its first road trip. I had an hour of non congested road to try out AP. …

I don’t know why folks are hung up on the exact words. The intent of what is being said is important not whether if it was exactly 60 minutes or 65.

I didn’t have a problem with the responses.

 

[S4WRXTTCS]

And again, Where in his paragraph did he note 60 Minutes! That was made up in your head like most of the facts you present.

I dunno about anyone else, but if someone starts out with "I had an hour of non congested road to try out AP." I think its safe to assume they had roughly 60 minutes to try it.

Sure there is some wiggle room with the non-congested roads aspect so its not concrete, but safe enough for our needs.

Did the person experience PB? Yes
Did the Phantom braking happen excessively during the duration of the test? Yes
What percentage of the Phantom braking was severe? unknown from the data provided.
Was it FSD Beta? No
Was it Tesla Vision? Yes
Undivided and divided roads? Unknown
Was any of it AEB? Doesn't look like it
What percentage was due to maps issue? Unknown.
What was the FCW set to? Unknown
Was was the following distance set to? Unknown

 

[Dewg]

Could I buy another vehicle that has 1, 2, 3, and 5? while having a good chance of no PB? It's a strong possibility. Here is the first google search I did, but there are probably a lot more.

Reading that article, looks like it's still just forward looking sensors (likely radar). We don't know how Telsa's TACC is using sensor data. We know that AP uses all sensors to know what's happening all around the car. Does TACC still use some parts of AP, such as all sensors? Could PB just be confusion from some of the sensors when there's overlap? We know that PB with Vision+Radar was usually due to discrepancies between the two systems. If Elon is to be believed, the radar was low resolution and would sometimes give faulty telemetry (signal bouncing from an overhead sign or bridge), which would conflict with vision saying there's nothing there. The car would have to err on the side of caution and assume there's something there, slowing you down.

Perhaps TACC would be better if it shutdown, or ignored sensor data from anything other than the forward cameras.

How would other cars with basic adaptive cruise control handle a car/truck on the side or in the blind spot coming/drifting into your lane (potentially causing an accident)? I know they can see the car ahead slowing down, but what about all the cars around you?

 

[S4WRXTTCS]

Reading that article, looks like it's still just forward looking sensors (likely radar). We don't know how Telsa's TACC is using sensor data. We know that AP uses all sensors to know what's happening all around the car. Does TACC still use some parts of AP, such as all sensors? Could PB just be confusion from some of the sensors when there's overlap? We know that PB with Vision+Radar was usually due to discrepancies between the two systems. If Elon is to be believed, the radar was low resolution and would sometimes give faulty telemetry (signal bouncing from an overhead sign or bridge), which would conflict with vision saying there's nothing there. The car would have to err on the side of caution and assume there's something there, slowing you down.

Perhaps TACC would be better if it shutdown, or ignored sensor data from anything other than the forward cameras.

How would other cars with basic adaptive cruise control handle a car/truck on the side or in the blind spot coming/drifting into your lane (potentially causing an accident)? I know they can see the car ahead slowing down, but what about all the cars around you?

TACC with Tesla Vision is exactly that.

It only uses the cameras, and the radar is disabled.

It's supposed to reduce PB due to overhead roads, signs, etc. It's also supposed to reduce phantom braking due to semi's close to the lane edge, but that was significantly reduced before I got Tesla Vision. If anyone drives on I5 between Seattle and Portland they know its practically owned by Semi's so you'll pass hundreds of them which gives ample opportunity to see how a vehicle responds to semi's on the lane or slightly over on occasion.

Tesla Vision does seem to introduce phantom braking due to other issues. It's also likely more prone to sun light angles, and weather events. I think this contributes to a lot of the differences in behavior we see.

I personally question the accuracy of Tesla Vision at long ranges at night.

A couple weeks ago I was driving home from Portland, and tried NoA as I hadn't tried it in awhile. What surprised me was it kept getting into the passing lane, and then getting back out of the passing lane. It did it around 5 times within a few minutes without ever passing the vehicle way in front of us. After the 5th time I turned off "exit passing lane", and then it decided to laugh at me by getting in the far right lane. So at least it has a sense of humor I guess. I turned it off after that stunt.

Now it could have just been some weird algorithm issue, but its as if the car couldn't really tell the speed of the vehicle way in front. This is something Radar does a really good job of. In fact Vision+Radar was a better predictor of closing speed than I was.

Now I've never been a fan of NoA due to poor lane logic so I don't have a lot of seat experience with it. But, I was surprised by the poor performance of Tesla Vision at night (auto brights turned off).

The other indicator that Tesla Vision might not really be there yet is the fact that its still limited in ways Vision+Radar isn't. Like Tesla Vision has a speed setting limit of 80mph versus 90mph.

As to other vehicles its hard to say without extensively testing them, and seeing how they react to certain situations. Things like vehicles up ahead getting close to lane line or slightly over. Some of them have accident avoidance and some of them don't.

Odds are the vehicles have more than adaptive cruise control. Regardless of the vehicle I would never count an honest accidental avoidance braking as a PB even if I felt like it was being a little nervous.