Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Phantom braking still an issue

This site may earn commission on affiliate links.
Ye I saw that one. Looks like it thought something was in front and did a full emergency stop on the motorway.

In the manual TACC is meant to only slow down for traffic in front, but otherwise not adjust the speed as speed limits change. However, I've read a lot of people have had it change speed anyway. I would be curious to know if you disabled collision warning / avoidance and used 'basic' TACC - would it still phantom brake?

The review seemed a little dramatic to me so I skipped to the end, but that could be because I find him generally a little tedious but hey. Each to their own.

My understanding was TACC wouldn’t phantom break but who knows.
 
Click the right scroll wheel and say "Bug report" for phantom braking. Tesla's AI will thank you. No assurance that it will ever get in front of a human but, hey, every little helps.

Coming off the M1 to the M25 my M3 with FSD slows for the curved ramp... YMMV
 
My experience is that when on TACC no AP don’t have FSD. 2020.28.6. So when merging from lane 3 to Lane 2 if I am merge next to Large Van or HGV the brakes WILL come on it’s very consistent. Learnt to have my foot on the go pedal. I now merge a bit later so the side camera does not see the vehicle inside me. Must Remember to bug report
 
  • Like
Reactions: Chickenlips
I had 4 incidents of phantom braking on my recent trip from South Wales up to the Scottish Borders then Highlands.

- All on the motorway (M6)
- All when using TACC (I rarely bother with Auto pilot)
- 3 were when approaching either overhead bridges or large overhead signs
- 1 was because the Nav thought the speed limit changed from 70 to 30. It hadn't of course.

In all cases I was able to quickly tap the accelerator and keep moving at pace so as not to put the fear of god into the cars behind me. Gave me a bit of a fright each time though, it's not a pleasant experience.
 
  • Like
Reactions: SalisburySam
Hoping that the autopilot rewrite will help reduce phantom braking.

I'm sure some events is the car overreacting to a scenario such as another vehicle (especially large) briefly drifting close to lane marking or TACC thinking you are going to accelerate into car infront when going to overtake etc.

Other scenarios are more wtf moments with no obvious rational cause such as gantries etc.

At the moment, AP/TACC uses a moment in time determination so its got little or no history running up to the event.

Hopefully, when the rewrite adds time to the mix, its going to have more pertinent information to go on, so for example the bridge that you have been approaching for several seconds that momentarily gets identified as a truck across the carriageway is less likely to cause the car to emergency brake.

I still think though that other phantom brake incidents will not be sorted, such as phantom braking for an intersection that you are passing beneath. Hopefully the map update will go some way in helping.

imho
 
Speed changes in TACC mode in response to real or perceived changes in speed limit definitely do occur, both down and up. However, it’s very unclear to me when they happen. In Autosteer mode they seem to happen more often, but always after going through the start of the lower speed limit rather than anticipating it.
 
I drove in the bad weather on Tuesday. Phantom braking happened a lot on the motorways. Couple of times it was under bridges or a gantry but a lot of cases were when near lorries. I think the spray from them makes them look much bigger than they are to the autopilot. Also, the visualisation kept showing them jumping into the lane I was in and back again, so I took to overtaking really wide in the outside lane.
 
when the rewrite adds time to the mix

It amazes me that they tried fsd without 'history'. Kind of like 'spot the ball' or 'guess what happens next' based on a single frame image.

Couple of times it was under bridges or a gantry but a lot of cases were when near lorries.

Certainly my two main triggers. I wonder if it is possible to get secondary radar returns (after a double reflection). Some of Green's videos show clusters of returns apparently directly in front of the car as it passes under bridges. The returns must come from something but evidently not from a real object actually in front of the car. I suspect large flat surfaces in specific positions relative to the car could be canditates for causing secondary reflection if such a thing is possible.
 
Last edited:
  • Like
Reactions: MrBadger
Is anyone worried like me that the rewrite will take months to bed in and debug? Presumably it cannot be a full rewrite or they would have to junk all the accumulated training by the neural networks. Why do people think the transition will be seamless?

In terms of ML (machine learning) models, training is always thrown away between new models. The way ML works to train a model is a load of input is provided to the system, and it then goes through a very fancy random number generator, finally there's an output. The output is then checked to ensure it identified the input correctly. For example if you put a photo of '3' into the model, did '3' come out? If it didn't the numbers in the middle are adjusted in a somewhat intelligent, yet also somewhat random way until the model starts to consistently say '3'.

Once this process is complete, the model is 'trained' and will be deployed out in the field to categorise real things. What will happen is a different input (photo) will be encountered and a '3' is interpreted as a '4' and so the model will need to be re-trained. This means basically starting from the beginning again, with some tweaks to both the training data (pre-defined input with known desired outputs) as well as how the numbers are calculated and adjusted in the middle.

The problem with Tesla right now, is due to hardware constaints (HW2/2.5) they can't process enough data and models to do it 'properly'. So the system looks, approximately every 9ms, and almost entirely just by the front main camera, at the whole scene in front of the car. It then makes all judgement calls about if there's a hazard in front, if the car needs to steer left or right, accelerate, etc. Then 9ms later it looks again completely fresh and decides again what to do.

Obviously looking at frames individually means it'll be both entirely reactive, and not smooth at all. So what Tesla have done is write loads and loads of code on top of the decisions / categorisations it makes to smooth them out. So 'emergency brake' becomes 'slow down the car with vigor' and 'hard left' becomes 'turn the wheel to the left more'.

The next version on the other hand is doing things how they should have been done to start with. First it looks at all the cameras so it has a mostly 360 degree view. This means things won't suddenly disappear / appear in the field of view causing panic, except for when camera view is blocked by weather. Next, the model is no longer looking at each frame fresh, but it looks at the progression of things. This means if an object suddenly appears it can see the movement of the object and whether it'll cross the car's path or is instead moving away or just fixed with the environment.

The other change is Tesla's internal development is significantly more robust than it was with AP2. AP2 to me looks like it was rushed because of a falling out with Mobileye. AP3 has a proper test suite where the inputs (video/radar) are played back and the system is verified that it does what it's meant to do.

Now, the end result is that yes, there will be some issues still sadly. New stuff is always less predictable than old stuff. But going forward after the release you shouldn't get anywhere near the number of regressions that you get now and it should just slowly improve as the edge cases are tested for explicitly. Of course in my opinion they still won't really get there without stereoscopic cameras and lidar but I'm not an expert. Also not providing basic cruise and adaptive cruise is shocking. The software is buggy, provide a basic fallback.
 
Once this process is complete, the model is 'trained' and will be deployed out in the field to categorise real things. What will happen is a different input (photo) will be encountered and a '3' is interpreted as a '4' and so the model will need to be re-trained. This means basically starting from the beginning again, with some tweaks to both the training data (pre-defined input with known desired outputs) as well as how the numbers are calculated and adjusted in the middle.
The bit I struggle with is how does the car know that the 3 is being wrongly interpreted other than there not being a crash or the driver pressing the accelerator again. My frustration is the learning doesn't happen until the new software version and for your particular complaint, most likely will not be fixed. For me the most obvious example is when a road diverges from one lane to two, A303 does this many times. Every time the car veers to the right lane and not the left, I hold the wheel and defeat AP and I wait to see it repeat it again the next day, groundhog style. It's not really a complaint, the car tech is stunning, way beyond the competition and its exciting to be literally along for the ride, easy to forget how privileged we are.
 
  • Like
Reactions: SalisburySam
Every time the car veers to the right lane and not the left, I hold the wheel and defeat AP and I wait to see it repeat it again the next day,

That's something I find a bit depressing. It feels as though little if anything changes on my regular routes despite the car uploading disengagement info back to Tesla. I suspect the truth is that with little evidence of appetite for embracing self drive testing in our regulatory framework, there isn't too much point Tesla focussing on UK at the moment.
 
  • Like
Reactions: mpandrew
I've just finished a 1060 mile run down the UK and back and I think the phantom braking is much improved. 90% of the time just a gentle slowing only twice today on the Northbound run did we have severe braking and none on the Southbound run on Friday.
 
Sudden braking phantom braking 10 times in 1000km drive in Australia. Occurs when overtaking long trucks an some cars. Various situations.
Very dangerous as trailing car when overtaking another is sometimes close to my tail. Speed drops from 110kmh to 80.
 
I thought this problem might have gone away with fewer reports but no. TACC disengaged with fairly sudden braking down from about 70 to 50 before I caught it as I passed under a bridge on the M40 last night. And gentler braking passing a curtain slider. Now on 2020.48.30. ....bad
 
Recently did two 350 mile drives - only had a couple of minor phantom braking events that self-resolved quickly. Generally performed well. One tip I have is every time you stop give all of the cameras a wipe with a cloth, seems to help reduced warnings at least.

I do wish they'd sort this issue, that said I still maintain autopilot has changed my life. My regular whole-length-of-England drives used to be exhausting, the Tesla has made it very easy and I arrive feeling much fresher. Furthermore I think the forced breaks for charging help as in the past I've thought I would just soldier on, which is not safe and not good for my bladder or back!