Phantom Braking getting worse - any ideas on how to stop it?

[texas_star_TM3]

Better still might be to sell your Tesla and purchase something that doesn't infuriate you so badly. Life is too short for that.

Good luck in whatever you decide.

so Tesla software deleting installed radar units and forcing customers on something as half-baked as "vision only" is no biggie... got it. The Model 3 was perfect before they forced it on vision only. And yes - I would currently not purchase a Tesla again if cruise-control doesn't work reliably like other OEMs. Not to mention parking sensors gone overnight with "vision only" will handle that as well. eventually. maybe.

My guess is that the facelift/ project highland will add a new HD radar unit as well as cameras in the front bumper and surprise... both issues are resolved. we all know that "vision only" was primarily to circumvent part shortages during covid / cost cutting. it's up to the fanboys to spin it though and declare nothing is there to see...or "buy something else" (which customers eventually will do and thank for supercharger network you can while still benefitting from a far superior charging system in scope and design).

 

[Supcom]

so Tesla software deleting installed radar units and forcing customers on something as half-baked as "vision only" is no biggie... got it.

I did not intend to imply that. My point was that if your car makes you so mad, you should find one that you will enjoy. It seems unlikely that Tesla will ever reenable your radar. so if it's that important to you, then you should seriously consider a different vehicle. That's what I would do.

 

[texas_star_TM3]

I did not intend to imply that. My point was that if your car makes you so mad, you should find one that you will enjoy. It seems unlikely that Tesla will ever reenable your radar. so if it's that important to you, then you should seriously consider a different vehicle. That's what I would do.

yeah... I will wait until late 2024 and make a decision then. if Model 3/Y Highland facelift include a radar or have fixed it otherwise... I'm still team Tesla. If not ... well... Ford and GM (+ Rivian / Polestar / Volvo as well ?) will start delivering 2025 model year EVs with NACS plug already built in and using SCs won't even require an adapter...

 

[Ridgeline]

maybe not. But i want Tesla to at least be confronted with it and have it on record as a service log. Again - the performance of TACC (not even talking AP) is INSANE on hot days on long stretches of road with mirages.

If i cant use cruise control going pretty much just under 5 hours straight on a well built out highway with 2 lanes each direction - separated.... when can I use cruise control then? No other car with TACC struggles this much and brakes repeatedly this hard for *nothing*.

and the kicker: I drive this route pretty much annually on our way to Taos and *never* had this issue before they made my radar equipped car "vision only".

for now it's going to be back into the Mazda CX5 for long roadtrips in spring/summer/fall.... apparently TACC in the Model 3 only works reliably when the asphalt isn't hot (zero issues on our drive down to Galveston for NYE a few months back). The feeling out of the sudden going from 80mph+ down to 50mph would freak anyone out... including other cars honking at you.

I can understand the logic of taking out radar because having radar and vision together in the control system complicates its software. Now I will pull from my knowledge of safety systems in the hydrocarbon processing industry. There are two independent systems one for controlling the process and one to manage the safety critical elements of the process. There was a separate set of sensors and actuators for the safety functions that were completely isolated and separate from the process control sensors and actuators. The safety system would override the process control system in the following way: For example if the temperature for some reason got dangerously too high, maybe a failure of a temperature sensor or whatever. The safety system's temperature sensor would trigger the closing or opening of a valve or whatever to lower the temperature. So lets see how that would work in a vehicle: The vision system sees what it thinks is something that needs application of hard braking, but before it actually slams on brakes it would check with the radar system to determine if something is really there (happens within milliseconds or microseconds). Thats a simple boolean logic function or "if then else" statement in software. Well.... that is not exactly like the two systems that I've described above except that it does keep the radar stuff isolated from the normal driving software controls thus not complicating it. Seems too simple and perhaps Tesla engineers have already tried it. I agree it doesn't fit "the best part is no part" ideology but it would be better IMHO than having the Fed (NHTSA) involved which is being discussed here multiple times.

 

[texas_star_TM3]

I can understand the logic of taking out radar because having radar and vision together in the control system complicates its software. Now I will pull from my knowledge of safety systems in the hydrocarbon processing industry. There are two independent systems one for controlling the process and one to manage the safety critical elements of the process. There was a separate set of sensors and actuators for the safety functions that were completely isolated and separate from the process control sensors and actuators. The safety system would override the process control system in the following way: For example if the temperature for some reason got dangerously too high, maybe a failure of a temperature sensor or whatever. The safety system's temperature sensor would trigger the closing or opening of a valve or whatever to lower the temperature. So lets see how that would work in a vehicle: The vision system sees what it thinks is something that needs application of hard braking, but before it actually slams on brakes it would check with the radar system to determine if something is really there (happens within milliseconds or microseconds). Thats a simple boolean logic function or "if then else" statement in software. Well.... that is not exactly like the two systems that I've described above except that it does keep the radar stuff isolated from the normal driving software controls thus not complicating it. Seems too simple and perhaps Tesla engineers have already tried it. I agree it doesn't fit "the best part is no part" ideology but it would be better IMHO than having the Fed (NHTSA) involved which is being discussed here multiple times.

100% agree...

relying ONLY on 1.2 MP cameras facing forward - through a windshield which might be dirty to some extent, glarring sunlight and mirages/ reflections over the road - to determine if an object requiring braking is there or not is incredibly non-redundant and actually worse than human eyes with a brain.

radar isn't fooled by mirages over the road and at least with vision + radar ... radar could act as the ultimate yes/no before applying the brakes. even crappy radars are really accurate within 100 meters and that's well within the stopping distance even at 80 mph...

 

[izzzyyd]

Same started noticing when I came down south Florida Texas Arizona

 

[David__]

Yes, Tesla is putting a design ideology ("has to be vision because people do it") above using whatever is the safest (radar) with current technology. Customers are getting an inferior dynamic cruise control experience vs other manufacturers as a result.

The most important question is "Is the car going to hit what is in front of it?". That's what radar answers the best. The inference based vision system is primarily trying to identify what the object in front is first, and then after maybe identifying it, guessing about how close it is based on training. That's super dumb and less safe then just making sure the car is not going to hit an object in front, regardless of what it might be. The way Tesla is doing dynamic cruise is very handicapped.

 

[finman100]

Model 3, 2018 Long range rear-drive. NEVER had basic autopilot issues in Nevada. been going from OR to AZ twice a year for 5 years. so...what's mine got that others do not?

 

[texas_star_TM3]

Model 3, 2018 Long range rear-drive. NEVER had basic autopilot issues in Nevada. been going from OR to AZ twice a year for 5 years. so...what's mine got that others do not?

your car isn't HW3 ... so your radar is still active... my July 2019 Tesla (HW3) with radar had the radar software deleted in August 2022 when it was force migrated to "vision only" which requires HW3+

 

[izzzyyd]

Model 3, 2018 Long range rear-drive. NEVER had basic autopilot issues in Nevada. been going from OR to AZ twice a year for 5 years. so...what's mine got that others do not?

People really need to stop speaking in absolutes it’s really tiring. When there are a handful of differences between hardware software and auto pilot versions….

 

[Supcom]

Yes, Tesla is putting a design ideology ("has to be vision because people do it") above using whatever is the safest (radar) with current technology. Customers are getting an inferior dynamic cruise control experience vs other manufacturers as a result.

The most important question is "Is the car going to hit what is in front of it?". That's what radar answers the best. The inference based vision system is primarily trying to identify what the object in front is first, and then after maybe identifying it, guessing about how close it is based on training. That's super dumb and less safe then just making sure the car is not going to hit an object in front, regardless of what it might be. The way Tesla is doing dynamic cruise is very handicapped.

I don't think that occupancy networks function the way you describe. Objects do not need to be identified to be added to the non-drivable space around the car.

 

[finman100]

plenty of phantom braking on I-90 thru Wyoming and Montana the past 5 years for me. weird.

 

[enemji]

I have had PB happen only on long empty stretches of road. With that in mind, this is why I believe it happens:

Tesla Vision is processing every frame and identifying what is ahead of it. This is happening at 36 frames/sec. Now imagine that you have 360 frames (10 seconds) that have no change of data. At that point, the FSD is probably designed to take a safe route of assuming that the vision is frozen, and hence it takes the decision to brake.

That being said, why is there no message on the screen indicating that is what happened. That is why it is still a theory.

 

[Supcom]

I have had PB happen only on long empty stretches of road. With that in mind, this is why I believe it happens:

Tesla Vision is processing every frame and identifying what is ahead of it. This is happening at 36 frames/sec. Now imagine that you have 360 frames (10 seconds) that have no change of data. At that point, the FSD is probably designed to take a safe route of assuming that the vision is frozen, and hence it takes the decision to brake.

That being said, why is there no message on the screen indicating that is what happened. That is why it is still a theory.

There are no two frames with no change of data. If nothing else, the dashed lane separator will be different. But, despite your seeing an unchanging landscape, the video has many changes from frame to frame.

 

[David__]

There was at least one accident where a tesla crashed into/under a semi-trailer which was perpendicular to, and crossing, the road. I'm betting there were several other accidents which were similar in that the crash was caused by autopilot failing to "identify" the object, so it hit it. That is the problem with the stupid identification-first based system. It has to first identify a pedestrian/car/whatever and only _then_ will it try to guess how far away it is. The key point is that if there is an object that the car is approaching, that the inference-based vision system doesn't recognize as anything, the car will just hit it.

A far better system is to first answer the question of, "is the car on track to hit an object ahead?" That's what radar will tell you. Who cares what the object is, we just care about not running into something. The inference-based vision system has to first try to identify something, because that is the only way it can then move on to the second priority of guessing how far away it is (using its ridiculously basic mono-camera in the front windshield).

 

[derotam]

I have had PB happen only on long empty stretches of road. With that in mind, this is why I believe it happens:

Tesla Vision is processing every frame and identifying what is ahead of it. This is happening at 36 frames/sec. Now imagine that you have 360 frames (10 seconds) that have no change of data. At that point, the FSD is probably designed to take a safe route of assuming that the vision is frozen, and hence it takes the decision to brake.

That being said, why is there no message on the screen indicating that is what happened. That is why it is still a theory.

Because of your statement about no message being displayed, I would say that your theory is not viable as a PB theory. Your theory can actually be tested. Empty street, piece of cardboard on a stick stuck out window and around to cover windshield cameras while on AP... I would theorize that there would be a display about taking over immediately due to camera being blocked, or something similar.

 

[Sambub]

Phantom braking seems to have improved considerably in my Dec build HW3 MYP, at least on local hwys in MN. I haven’t had an incident in several months.

We have a road trip planned to CO upcoming so wondering how that will go.

 

[texas_star_TM3]

Phantom braking seems to have improved considerably in my Dec build HW3 MYP, at least on local hwys in MN. I haven’t had an incident in several months.

We have a road trip planned to CO upcoming so wondering how that will go.

you need long stretches of open road with plenty of mirages over the hot asphalt. standard driving conditions in West Texas from April - October. Not so sure about Minnesota where the daytime highs even in peak summer rarely exceed the nighttime lows in Texas...

 

[enemji]

Because of your statement about no message being displayed, I would say that your theory is not viable as a PB theory. Your theory can actually be tested. Empty street, piece of cardboard on a stick stuck out window and around to cover windshield cameras while on AP... I would theorize that there would be a display about taking over immediately due to camera being blocked, or something similar.

It is still viable as a theory. If the message did show up, it would prove it right or wrong.

 

[enemji]

Phantom braking seems to have improved considerably in my Dec build HW3 MYP, at least on local hwys in MN. I haven’t had an incident in several months.

We have a road trip planned to CO upcoming so wondering how that will go.

CO is a horrible state. The same highway has multiple speed limits and everyone drives at 70mph. Tesla keeps slowing doen from 65 to 55 to 35.