Phantom braking so bad I want to return my car

[Fourdoor]

Nobody I have seen in this thread is claiming that any other manufacturer is a "Tesla killer"... we all want Tesla to thrive. But keeping the blinders on about other companies with lane centering and adaptive cruise control systems that work properly, and pretending that Tesla's system is as good as (let alone better than) other manufacturers will hurt Tesla in the long run. Tesla is pursuing edge case self driving while ignoring basic adaptive cruise control and lane centering. This isn't customer driven, it is an obsession of the founder of the company with massive amounts of ego invested in being right about not needing anything other than cameras, and being the first to have real self driving cars.

As to your computer industry analogy, look at the flip side. Computers (cars) have existed for a century and then a small innovative company comes alone that makes them better, faster, and more efficient. Do the existing computer companies go belly up and die? Or do they adapt? Tesla will never be "killed" by the legacy auto makers... but in reality Tesla will not "kill" the legacy auto makers either. It most likely will become just another car maker amongst the crowd.

Keith

Well, geezersquid disagrees with me. I guess he DOES think the other auto makers have Tesla killer levels of autonomy.

Keith

 

[TheTheocracy]

The sensor-driven vs learning-driven system is always a topic for debate, as many researchers and companies try to give the most robust solution in the quickest time possible.

I think multi-modal sensor platforms are always a quicker, more robust solution initially, but they have their own set of problems (sensor data fusion and decision making, data for training algorithms, cost of hardware, etc.). LiDAR processing, for example, is time consuming and requires much more hardware, depending on the algorithms you use (pointcloud processing, 3D voxels). Even some sensors are overkill for an application. Do you really need centimeter precision of a car in front of you to determine when to brake?

Strict 2D vision system are much harder to create algorithms for, but the push for deep learning has allowed better object detection and image segmentation of the data. One thing that plagues 2D deep learning is adversarial examples and instantaneous false positives. If you have done any degree of deep learning object detection in 2D, you will most likely see random false positive events at various frames in video, but not usually across multiple frames. It's probably why there are lot of phantom braking events. With that, I don't think making a more robust vision only obstacle avoidance and TACC is that simple to do.

This doesn't even mention route planning, space-time analysis, use of Recurrent NNs, things beyond just a sensor set.

In any case, I admire Tesla going with a vision only approach. There really should be a way to get all the information you need with a set of cameras and algorithms. As many have said, it will probably be a little more inferior in some cases until it can deal with those errors. I find myself turning off TACC/Autopilot or hover over the accelerator because of it.

Sidenote: I think the quantitative error displayed with Beta updates may be a little misleading. 15% improvement on error is 15% on the original error. For example, object detection has 15% error, an update improved error by 15%, meaning error is now at 12.5%. Percent improvement doesn't show error itself, only it's improvement, which may or may not translate to user experience (usually doesn't).

 

[stopcrazypp]

You have never seen a Tesla radar equipped car where the driver was paying attention run into the back of a fire truck either. People wanting the car to "do it all" while they pay no attention are the problem. I want a car that centers itself in the lane of travel and maintains speed in that lane unless the car in front of me slows down. I want it to do this while I look out the windshield / windows monitoring for dangers that are not directly ahead of me. Wanting the car to monitor things outside of those parameters is what causes the problems. Oncoming traffic in the other lane? I want the car to ignore that. Speed limit signs or lack of signage? I want the car to ignore that. On rural highways you may not see a speed limit sign for 15 or more miles... if a Tesla travels more than 5 miles without seeing a speed limit sign it defaults to 45 mph speed limit... in my area by law a rural highway without speed limit signs by default has a speed limit of 65 mph. The car should not be able to arbitrarily limit the speed the owner of the car can select on the cruise control system. A 15 mph differential between what I can legally drive on a road and what the car will allow me to set the cruise control too (5 mph higher than it thinks the speed limit is) is ridiculous. I think the move to vision only was a mistake that will be mitigated by software improvements. If Tesla just said "screw it, here is adaptive cruise and lane centering and that is all you get on basic auto pilot" they could implement that and it would be virtually perfect within weeks, possibly days in the vision only cars. Trying to teach the car to save everyone's ass when they drive without paying any attention will take years. That is fine, but it should be limited to the FSD cars, while the rest of us get functional cruise control and lane centering.

Keith

The thing is Tesla has NHTSA breathing down their necks, and there's been a bunch of widely reported accidents that involved AP where the car failed to respond to those things and it had led to fatal results. Even though phantom braking is annoying and may be a deal breaker in terms of use, it's quite low in terms of risk of a fatal accident, even lower if Tesla implements some checks for vehicles coming from behind (especially if braking/slowing is relatively mild, unlike AEB that may bring things to a complete stop at the extreme). Tesla is banking on eventually tamping down on the false positives while maintaining the same safety checks (as they have done in the past through the various AP hardware changes), but it'll be a very long process.

There are some easy things Tesla can do like put toggles for ACC/AP to ignore certain stimuli (which I'm fairly certain other systems do anyways already), but I think the lawyers would say no. Because even if they put a clear warning on the toggle about the risks of using it, they would likely be sued for it anyways if anything happens and NHTSA would likely frown on it (just like the whole passenger gaming thing that an investigation was recently opened for).

 

[jama999]

Yep. The FSD Beta 10.7 release notes show that it appears that they have made significant progress on the "false slowdowns":

Now the question is when will those updates make it in to the public Tesla Vision releases?

So, this is really interesting and the first acknowledgement from Tesla that this is actually a problem (that I know of?) - As Ive said before, it feels to me like FSD is the primary thread of development and attention from the software folks at Tesla right now, and the standard 'table steaks' features are just being maintained but not really improved or patched.

I should also state again - in my car, this is NOT happening on corner case 2 lane roads with traffic close to me - its happening on big, wide 4 and 5 lane roads with sparse traffic at speed. Yesterday was the first time I drove my car since the last SW update (2021.44.6) - I think, and I was wondering if there might be a patch in there for it. But, no - about 2 miles from my home on a completely stretch of open road on SH71, I had the standard TACC set at 60. On a slight curve with me in the right lane with no one in front or behind me, the car slowed from 60 down to 39 then made an alert chime. There was nothing else in the road, the sun was up but it was midday, no other indications as to why it suddenly decided there was a problem. I will actually post a video if I can get one, seems to do it in the same places every time.

So, I still maintain that it would be great if Tesla would acknowledge this to the rest of us non-coolkid-FSD beta types and just give us a "dumb" cruise control until they are able to sufficiently patch this.

 

[drtimhill]

Do the existing computer companies go belly up and die? Or do they adapt? Tesla will never be "killed" by the legacy auto makers... but in reality Tesla will not "kill" the legacy auto makers either. It most likely will become just another car maker amongst the crowd.

While my crystal ball is no clearer than anyone else, actually in the computer company analogy they DID all get killed. The "big" makers, huge companies like IBM, Digital, Tandem, Cray etc are either extinct or a mere shadow of what they used to be. There is a reason they use the word "disruptive" for major technology shifts.

And I'm not saying this will happen, no-one knows of course . look at Apple .. virtually extinct and on their knees at one point and now a titan. But I still think is more than just "drop out ICE engine, drop in a battery .. done!" that many people seem to assume. Either way, it's going to be fun watching as things evolve

 

[Fourdoor]

The thing is Tesla has NHTSA breathing down their necks, and there's been a bunch of widely reported accidents that involved AP where the car failed to respond to those things and it had led to fatal results. Even though phantom braking is annoying and may be a deal breaker in terms of use, it's quite low in terms of risk of a fatal accident, even lower if Tesla implements some checks for vehicles coming from behind (especially if braking/slowing is relatively mild, unlike AEB that may bring things to a complete stop at the extreme). Tesla is banking on eventually tamping down on the false positives while maintaining the same safety checks (as they have done in the past through the various AP hardware changes), but it'll be a very long process.

There are some easy things Tesla can do like put toggles for ACC/AP to ignore certain stimuli (which I'm fairly certain other systems do anyways already), but I think the lawyers would say no. Because even if they put a clear warning on the toggle about the risks of using it, they would likely be sued for it anyways if anything happens and NHTSA would likely frown on it (just like the whole passenger gaming thing that an investigation was recently opened for).

The problem is Tesla didn't market its system as adaptive cruise control and lane centering, if they had then the NHTSA wouldn't have a leg to stand on. Nobody expects a Chevy Silverado with Super Cruise to avoid all accidents while the driver drifts off to sleep, or plays games on their phone... but people DO expect that (incorrectly) from a Tesla... and it isn't anyone's fault but Tesla.

For example, a guy on another message board I frequent had a chance to drive his friends Tesla on winding mountain roads with switchback turns (25 mph)... the Tesla owner kept telling the guy driving the car to "trust the technology"... on one switchback they took the 25 mph curve at over 40 mph because the AP system ISN'T BUILT FOR THOSE CONDITIONS... but the owner of the car thinks it is, and was telling the guy driving the car that it was. If they had gone off the road into a ravine this dipshit would have been one of the idiots trying to file a lawsuit against Tesla and reporting them to the NHTSA.

Tesla as a company could have fully functional adaptive cruise control and lane centering if they could trust their customers to pay attention and handle the complex parts of driving. Instead they repeatedly tell the customer to pay attention (be vigilant for when you have to take control) while simultaneously trying to do the complex tasks of driving for the customer to prevent the driver from ever having to take control. All I want from them is a dividing line between basic AP (driver in control other than low end Level 2 control aka lane centering, and cruise control) and Full Self Drive where the car is capable of high end level 2 autonomy.

The lane centering is fantastic, it is just the speed control that sucks. I don't think I will ever get what I want, so my best hope is for less interference from the adaptive cruise control system. My greatest fear is that they will screw up the lane centering as bad as they have screwed up the speed control.

Keith

 

[jama999]

While my crystal ball is no clearer than anyone else, actually in the computer company analogy they DID all get killed. The "big" makers, huge companies like IBM, Digital, Tandem, Cray etc are either extinct or a mere shadow of what they used to be. There is a reason they use the word "disruptive" for major technology shifts.

And I'm not saying this will happen, no-one knows of course . look at Apple .. virtually extinct and on their knees at one point and now a titan. But I still think is more than just "drop out ICE engine, drop in a battery .. done!" that many people seem to assume. Either way, it's going to be fun watching as things evolve

Seems to me the cars that are taking the "drop out ICE engine, drop in a battery .. done!" are already a failure. I drove a Hyundai Kona electric, an electric Mini and a couple others recently to review for a friend's Youtube channel, and they all seem pretty unimpressive on a lot of different levels. I even think the Electric f150, which is pretty much a regular F150 with a gigantic 135kw battery stuffed between its existing ladder frame, is a pretty poor piece of engineering. The ones that are going to succeed - at least engineering wise - are going to be from the ground up EVs with a focus on the battery technology and the weight, not junky ICE cars that have been retrofitted.

Just my 2c,
J

 

[Fourdoor]

Either way, it's going to be fun watching as things evolve

My crystal ball is a bit cloudy as well, but I am enjoying the ride! First electric drivetrain for me was a first generation Chevy Volt, then when that was destroyed by a lightning strike to the building it was plugged into for charging I got a second generation Volt... then the Bolt (still have that one) and the MYP.

Keith

 

[AGARG]

While I'm not defending phantom braking, how sure are you those other cars you mention would have braked in a genuine emergency? Of course, we all want a car that brakes in a genuine emergency, but not otherwise, but there is no such thing. Would you prefer no phantom braking but a system that failed to brake in an emergency? How would you balance such a system?

These systems work flawlessly on existing automakes, I can attest to that. I have first hand experiened it on my Infinity, no phanton braking ever but two when someone lost control and crossed out lane on highway the car stopped us.
I also own a Mustang Mach E and have driven it about 10k miles and NO reports of phantom braking. Ford has not released the Blue Cruise update for my car but I have the intelligent Cruise with lane assist and have extensively used it on 2 lane and major highways. No phantom raking to report but auto steer and distance control needs some fine turning on Mach E. I have not yet received my MY, so remain to be seen. If the PB issue is as severe as people reported here, I might be keeping the Mach E over MY.

 

[Fourdoor]

These systems work flawlessly on existing automakes, I can attest to that. I have first hand experiened it on my Infinity, no phanton braking ever but two when someone lost control and crossed out lane on highway the car stopped us.
I also own a Mustang Mach E and have driven it about 10k miles and NO reports of phantom braking. Ford has not released the Blue Cruise update for my car but I have the intelligent Cruise with lane assist and have extensively used it on 2 lane and major highways. No phantom raking to report but auto steer and distance control needs some fine turning on Mach E. I have not yet received my MY, so remain to be seen. If the PB issue is as severe as people reported here, I might be keeping the Mach E over MY.

Dude, due to your post count and admitting that you don't yet own a Tesla in the minds of many on this board (not me) your opinion is not only worthless, it is intentional lies. Many can't see past their bias to accept that Ford makes something more advanced than the Model T, or that legacy auto makers are not putting their cars together with stone tools.

Keith

 

[roblab]

You will continue to make excuses and say, "it isn't meant for that use case" (false statement) until they finally fix it over the next year or two... then you will claim that it ALWAYS worked great in all use cases and the people who say otherwise are short sellers.

Keith

Nope. I have no excuses to make. Phantom BRAKING happens, but it's pretty well known where and when and why for, so it's fairly well known that Tesla has a lot of work to do before it's really ready for prime time. In the meantime, drivers should report these things to Tesla, including locations. Somewhere down the road they will figure out what causes it on Teslas, and what fixes it, and we will all be appreciative that it happens less and less often. MINE has always worked great, but I don't use TACC on back roads, and don't use TACC much at all, so there's that.

Long term Tesla owners realize that Tesla uses us all as Beta Testers.

 

[drtimhill]

Dude, due to your post count and admitting that you don't yet own a Tesla in the minds of many on this board (not me) your opinion is not only worthless, it is intentional lies. Many can't see past their bias to accept that Ford makes something more advanced than the Model T, or that legacy auto makers are not putting their cars together with stone tools.

It's neither worthless nor lies, its a data point, with whatever implicit biases he carries (like we all do, not denigrating the poster). The MachE comments are a good data point, though it should be noted that when Sandy Munro drove a mach-E on a freeway the autosteer could not even cope with a gradual curve (it disengaged). Of course the software might be updated now, but everyone has conformation bias (the poster noted that this as a minor issue, I wonder if he would be ok with this on a Tesla). The point of my questioning earlier wasn't to defend Tesla (they clearly need to do more work on FB, and they are), but to remind people that EVERY car from EVERY maker has good points and bad points.

 

[RDoc]

FWIW in the entire time we've owned our S I think we may have had somewhere between 5 and 15 phantom braking events total.

 

[stopcrazypp]

The problem is Tesla didn't market its system as adaptive cruise control and lane centering, if they had then the NHTSA wouldn't have a leg to stand on. Nobody expects a Chevy Silverado with Super Cruise to avoid all accidents while the driver drifts off to sleep, or plays games on their phone... but people DO expect that (incorrectly) from a Tesla... and it isn't anyone's fault but Tesla.

For example, a guy on another message board I frequent had a chance to drive his friends Tesla on winding mountain roads with switchback turns (25 mph)... the Tesla owner kept telling the guy driving the car to "trust the technology"... on one switchback they took the 25 mph curve at over 40 mph because the AP system ISN'T BUILT FOR THOSE CONDITIONS... but the owner of the car thinks it is, and was telling the guy driving the car that it was. If they had gone off the road into a ravine this dipshit would have been one of the idiots trying to file a lawsuit against Tesla and reporting them to the NHTSA.

Tesla as a company could have fully functional adaptive cruise control and lane centering if they could trust their customers to pay attention and handle the complex parts of driving. Instead they repeatedly tell the customer to pay attention (be vigilant for when you have to take control) while simultaneously trying to do the complex tasks of driving for the customer to prevent the driver from ever having to take control. All I want from them is a dividing line between basic AP (driver in control other than low end Level 2 control aka lane centering, and cruise control) and Full Self Drive where the car is capable of high end level 2 autonomy.

The lane centering is fantastic, it is just the speed control that sucks. I don't think I will ever get what I want, so my best hope is for less interference from the adaptive cruise control system. My greatest fear is that they will screw up the lane centering as bad as they have screwed up the speed control.

Keith

Even though people like to focus a lot on the marketing, NHTSA does not care about that part. As per the 2016 investigation, what they care about is if there is a potential for abuse and if the manufacturer is cognizant of it and actively working to fight against it and to handle things when such abuse happens.
https://static.nhtsa.gov/odi/inv/2016/INCLA-PE16007-7876.PDF
Even though nowadays people in some threads seem to be trashing AP, from back in those days, AP was loved because it was way better than competition in keeping the car in lane and keeping up with traffic, so much so that people may have trusted it too much (especially when back then the system was way less strict about keeping their hands on the steering wheel). I'm not really convinced people would love it nearly as much if the capabilities was pared down to be more basic, regardless of the marketing (which largely only matters for inexperienced users, which tend to naturally be more cautious first time using the system). Anyways, the cat is out of the bag already, and I doubt NHTSA would take it well if Tesla released an option that reduced the safety checks in the name of convenience at this point in time.

 

[TheLex]

I don't understand why Tesla folks like to rag on other manufacturers, as if they're inherently inferior. My F450 tows a 35 ft fifth wheel, and I've been all over the western USA and Canada. NO phantom braking, and yes, I have adaptive cruise on when I'm towing. We are 53 feet long and over 22k lbs weight. My truck's adaptive cruise works flawlessly. My Subaru Ascent has adaptive cruise and it works flawlessly and I've gone back and forth to Canada from NorCal with that vehicle multiple times. At this point adaptive cruise is an established technology that works for MANY other manufacturers without issue. Why Tesla can't get this part right and even refuses to acknowledge it is beyond me.

 

[turnbowm]

These systems work flawlessly on existing automakes, I can attest to that. I have first hand experiened it on my Infinity, no phanton braking ever but two when someone lost control and crossed out lane on highway the car stopped us.
I also own a Mustang Mach E and have driven it about 10k miles and NO reports of phantom braking. Ford has not released the Blue Cruise update for my car but I have the intelligent Cruise with lane assist and have extensively used it on 2 lane and major highways. No phantom raking to report but auto steer and distance control needs some fine turning on Mach E. I have not yet received my MY, so remain to be seen. If the PB issue is as severe as people reported here, I might be keeping the Mach E over MY.

Thanks for sharing your experience. I found your post to be informative and another data point for consideration.

 

[Pilo]

I have a MYP on order, soon to be delivered. My current daily driver is a 2017 Hyundai Ioniq Premium EV that I bought in 2020 with 35K miles for $16K. It has all of the driver assist stuff (lane warning but no autosteer). The smart cruise works perfectly. I've done 15K miles, with the smart cruise on probably for 25% or more of those miles. It's never done anything weird. I consider the way it works to be really really good. I'm sort of surprised to see this thread since I had no idea some people were having this issue. I can't add anything except that it will be disappointing to deal with this issue if I have it. I guess I'll find out soon enough.

 

[drtimhill]

I have a MYP on order, soon to be delivered. My current daily driver is a 2017 Hyundai Ioniq Premium EV that I bought in 2020 with 35K miles for $16K. It has all of the driver assist stuff (lane warning but no autosteer). The smart cruise works perfectly. I've done 15K miles, with the smart cruise on probably for 25% or more of those miles. It's never done anything weird. I consider the way it works to be really really good. I'm sort of surprised to see this thread since I had no idea some people were having this issue. I can't add anything except that it will be disappointing to deal with this issue if I have it. I guess I'll find out soon enough.

Please post your experiences, positive and negative, here when you get the car and have had time to try it out sufficiently.

 

[Pilo]

Will do

 

[ftmaybe]

I have a radar car that works. I have no severe phantom braking issues (or even mild ones really). I signed up for the FSD beta the day before yesterday. I have a 99 driving score. If I get into the beta, go “vision only” and I report major problems with phantom braking will this still be dismissed as marginal / not directly related to vision only? Or will it be assumed that I am now lying or exaggerating?


I am mainly doing this because I have a car on order and I want to know what I am getting into with vision only. But, at least I can also give some feedback here for others as well.

I got access to the FSD beta a few hours ago. I will test the adaptive cruise but it seems the real issues show up with sun and shadows. Given the persistent rain and fog we’ve been getting lately it might take a week or two for the sun to return to do any real tests.

However - this will be a real test. My radar car works great in all situations. Now that I am “vision only” in the FSD beta, we will see if my exact same “good” car goes downhill strictly with a firmware change.