The same thing for me a couple of times, it supposes to turn right even though the nav says that but stay on the left lane, 0.3, 0.2 miles keep going and then turn left instead.Had the same thing happen to us yesterday - wasn’t a familiar route, but the map had us taking a left, then a U turn in 2 blocks rather than simply taking a right.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD Beta 10.69
- Thread starter Buckminster
- Start date
Well, he specifically starts the video talking about what would happen if you replaced all cars on the road with Teslas WITHOUT steering wheels, and, based on current Autopliot safety statistics, how many lives would be saved. But the statistics don’t in any way indicate how safe Teslas without steering wheels would be - only how safe Teslas on AutoPilot WITH a driver behind the wheel are. That’s my point.
willow_hiller
Well-Known Member
Some anecdotal evidence that 10.69.25.2 now picks routes that are easier for FSD Beta:
thats smart. Sometimes it picks turn that even I don’t want to make.
sleepydoc
Well-Known Member
The braking is about how my mother used to drive. Not saying it’s good, just saying.
It’s too late/aggressive IMO, but in some parts of the country it’s probably typical.
AlanSubie4Life
Efficiency Obsessed Member
Actually this is often not the problem, as has been detailed in videos extensively (and exhaustively (exhaustingly???)).
Study the behavior carefully and compare to a somewhat fast human stop (with or without friction brakes).
Last edited:
When I enabled one pedal driving, I thought "I'm not gonna have to replace my brake pads the entire time I own this car," and I typically own my cars for ten years.
Then I became an FSDb tester.
Anyone know some good brake pads for a 2019 P3D-? Asking for a friend.
Then I became an FSDb tester.
Anyone know some good brake pads for a 2019 P3D-? Asking for a friend.
Welp my FSD subscription ends in a day. This was my first time ever experiencing FSDb and I have to say that I am impressed with it. Ignoring the promised timelines, FSDb in its current state is pretty good. It can do my normal commutes besides a U-Turn; even if it decides to make some weird decisions. It'll still get to the destination without me having to disengage. I've only had 30 days with it but it's nice to see what the current state of FSDb is from a first person experience.
Now to pay off all these "No Turn On Red" tickets...
Now to pay off all these "No Turn On Red" tickets...
You’re not kidding. It’s almost like they purposely made FSDb slam and use the brakes excessively so we can pay them $200/hour for labor to replace our pads with another set.
jebinc
Well-Known Member
IIRC, when using the radar, the following distance wad greater than the now vision rendition. Maybe a limitation of the vision code?
AlanSubie4Life
Efficiency Obsessed Member
No I don’t think so. But I am not talking about AP here. In FSD (which never had radar in release to customers), the following distance has no effect (nor does Chill/Aggressive driving style as far as I can tell)..
jebinc
Well-Known Member
Really? I thought early FSDb used the radar until the disabled it several updates ago. Perhaps I’m mistaken (again).
AlanSubie4Life
Efficiency Obsessed Member
When people were added to FSDb they received TeslaVision. (Can check the email.)
I initially thought this was related to the rumored new NN controlling vehicle dynamics but maybe the vehicle was somehow struggling to station keep at the downhill stop with throttle (reverse) versus using brake/hold. If so it's still very weird given EVs have regen and brake/hold and one wouldn't expect new features for a bug fix release. Stuff like this given an impression the team is out of control at times and willing to throw caution to the wind with s/w development.
Last edited:
Daniel in SD
(supervised)
V9 of FSD beta removed radar. That's when FSD Beta became "mind blowing."
Tesla FSD will be pure vision and not rely on radar use
Elon tweeted that v9 of the FSD beta would remove its reliance on radar completely and instead determine decisions based purely on vision. Humans don’t have
www.notateslaapp.com
AlanSubie4Life
Efficiency Obsessed Member
Exactly. I don’t include early beta recipients as customers (this is why I specified “in release to customers”) since they were able to effectively monetize it.V9 of FSD beta removed radar. That's when FSD Beta became "mind blowing."
Tesla FSD will be pure vision and not rely on radar use
Elon tweeted that v9 of the FSD beta would remove its reliance on radar completely and instead determine decisions based purely on vision. Humans don’t have
(Also that early version not relevant to following distance since no one knows, and no one experienced it.)
Last edited:
I haven't noticed an issue with following too close to a lead vehicle. However, excess and delayed braking are unquestionably typical. The issue seems to be FSDb's ADD-like focus on the lead vehicle resulting in FSDb's surprise of upcoming red light intersections, lines of cars stopped or slowed ahead, etc. Or for that matter, a myopic-like forward focus that leads to the same behavior with no lead vehicle. It's not how safe drivers operate a vehicle.
AlanSubie4Life
Efficiency Obsessed Member
Here's an example, following at 1.5 seconds at 45mph. This is far far too close and not what a safe driver would typically do (ignore the falsely reassuring GoPro perspective and just count the frames!).
This is an annoyance, but unfortunately it's not the main source of the problems. Increasing following distance would just allow slower braking if the person in front stops suddenly, better visibility, and it would feel more comfortable and be safer.
It’s partly this, but if you watch and feel carefully you’ll see that is not the primary problem.
Red lights are usually noticed in plenty of time. Though there are certainly exceptions!
The primary issue is that the initial brake application is too hard. Then, the car stops braking (20% of original force or so).
Then a short time later after coasting and not decelerating at all for a second or so, it of course needs to slow faster, so brakes harder.The car could come to a fast, consistent halt, with late braking for stopped traffic. Wouldn't be ideal for brake use and would be scary and non-optimal. Of course better to slow early enough. But right now it is usually slowing early enough for a quite sedate stop (maybe a little friction brakes but no big deal), but it just isn't doing it!!!
Here's an example. Sure, it was about 1-2 seconds later than I would have been easing off the accelerator to minimize slowing forces and guarantee regen only. But look at that regen bar!!!
Here is the first onset:
Then it goes to nothing (again, at most 20% of original force on this non-linear display - observe how the speed stops changing (but remember speedometer is delayed so is not synced with the regen)).
And it maintains that for a second or so. Here it is, still at it! Hardly lost any speed, notice (this is significantly downhill so you get regen from coasting at constant speed)!!!
Then it slams on the brakes again, having wasted an apparently sufficient amount of perfectly good stopping distance:
Again, it was SLIGHTLY late to start braking here. But literally I could have disengaged FSD with the stalk at the moment it started braking, removed my foot from the brake and accelerator, and let it coast to a near halt. Then, feather in the brakes at the end of the stop to ensure I did not hit the person in front. I definitely think I would have had to use the brakes, but it would have been very mild, in spite of the slight delay in reaction to stopped traffic, and it would have been just light application applied over a longer period, towards the end of the stop, at as low a speed as possible.
So the timing of the onset of braking is usually "ok" - this is an example. I'd LIKE it to start earlier of course. But that's NOT the main problem.
This same pattern occurs on nearly every problematic stop. No one understands why.
Here's where I discuss it in real time:
"It is completely insane."
Last edited:
Sorry but just as an observer I don’t follow nor view the issues you complain about. “Why did it slow down” ahhhh the car in front of you drastically slowed down, fairly self explanatory. “I would have gone for it on yellow” most including me Would Not! Not worth the gamble. “it’s slamming on the brakes!” Ummm going from 42 to 39 is not slamming on the brakes sorry. Your views are appreciated though most often exaggerated to an extreme in my view. “It’s following To close!” Yet a second later you disable and reduce that gap by near half. Still seems to me you are after perfection as compared to how You drive verses a standard universal safety like most other drivers. Just my view.
Here's an example, following at 1.5 seconds at 45mph. This is far far too close and not what a safe driver would typically do (ignore the falsely reassuring GoPro perspective and just count the frames!).
This is an annoyance, but unfortunately it's not the main source of the problems. Increasing following distance would just allow slower braking if the person in front stops suddenly, better visibility, and it would feel more comfortable and be safer.
It’s partly this, but if you watch and feel carefully you’ll see that is not the primary problem.
Red lights are usually noticed in plenty of time. Though there are certainly exceptions!
The primary issue is that the initial brake application is too hard. Then, the car stops braking (20% of original force or so).
The car could come to a fast, consistent halt, with late braking for stopped traffic. Wouldn't be ideal for brake use and would be scary and non-optimal. Of course better to slow early enough. But right now it is usually slowing early enough for a quite sedate stop (maybe a little friction brakes but no big deal), but it just isn't doing it!!!
Here's an example. Sure, it was about 1-2 seconds later than I would have been easing off the accelerator to minimize slowing forces and guarantee regen only. But look at that regen bar!!!
Here is the first onset:
View attachment 898439
Then it goes to nothing (again, at most 20% of original force on this non-linear display - observe how the speed stops changing (but remember speedometer is delayed so is not synced with the regen)).
View attachment 898440
And it maintains that for a second or so. Here it is, still at it! Hardly lost any speed, notice (this is significantly downhill so you get regen from coasting at constant speed)!!!
View attachment 898442
Then it slams on the brakes again, having wasted an apparently sufficient amount of perfectly good stopping distance:
View attachment 898443
Again, it was SLIGHTLY late to start braking here. But literally I could have disengaged FSD with the stalk at the moment it started braking, removed my foot from the brake and accelerator, and let it coast to a near halt. Then, feather in the brakes at the end of the stop to ensure I did not hit the person in front. I definitely think I would have had to use the brakes, but it would have been very mild, in spite of the slight delay in reaction to stopped traffic, and it would have been just light application applied over a longer period, towards the end of the stop, at as low a speed as possible.
So the timing of the onset of braking is usually "ok" - this is an example. I'd LIKE it to start earlier of course. But that's NOT the main problem.
This same pattern occurs on nearly every problematic stop. No one understands why.
Here's where I discuss it in real time:
"It is completely insane."
The fisheye-like lens distorts so it doesn't look too close but I tried counting one-1000, two -1000 and seemed to be about 2 secs. I agree FSDb braking action is an unnatural step-response no one in correct mind or senses would apply but, like I said, for me, over and over again, it's primarily related to myopic focused regions around the EGO. But there isn't just one or two issues here.
Here's an example, following at 1.5 seconds at 45mph. This is far far too close and not what a safe driver would typically do (ignore the falsely reassuring GoPro perspective and just count the frames!).
This is an annoyance, but unfortunately it's not the main source of the problems. Increasing following distance would just allow slower braking if the person in front stops suddenly, better visibility, and it would feel more comfortable and be safer.
It’s partly this, but if you watch and feel carefully you’ll see that is not the primary problem.
Red lights are usually noticed in plenty of time. Though there are certainly exceptions!
The primary issue is that the initial brake application is too hard. Then, the car stops braking (20% of original force or so).
The car could come to a fast, consistent halt, with late braking for stopped traffic. Wouldn't be ideal for brake use and would be scary and non-optimal. Of course better to slow early enough. But right now it is usually slowing early enough for a quite sedate stop (maybe a little friction brakes but no big deal), but it just isn't doing it!!!
Here's an example. Sure, it was about 1-2 seconds later than I would have been easing off the accelerator to minimize slowing forces and guarantee regen only. But look at that regen bar!!!
Here is the first onset:
View attachment 898439
Then it goes to nothing (again, at most 20% of original force on this non-linear display - observe how the speed stops changing (but remember speedometer is delayed so is not synced with the regen)).
View attachment 898440
And it maintains that for a second or so. Here it is, still at it! Hardly lost any speed, notice (this is significantly downhill so you get regen from coasting at constant speed)!!!
View attachment 898442
Then it slams on the brakes again, having wasted an apparently sufficient amount of perfectly good stopping distance:
View attachment 898443
Again, it was SLIGHTLY late to start braking here. But literally I could have disengaged FSD with the stalk at the moment it started braking, removed my foot from the brake and accelerator, and let it coast to a near halt. Then, feather in the brakes at the end of the stop to ensure I did not hit the person in front. I definitely think I would have had to use the brakes, but it would have been very mild, in spite of the slight delay in reaction to stopped traffic, and it would have been just light application applied over a longer period, towards the end of the stop, at as low a speed as possible.
So the onset of braking is usually "ok" - this is an example. I'd LIKE it to start earlier of course. But that's NOT the main problem.
This same pattern occurs on nearly every problematic stop. No one understands why.
Here's where I discuss it in real time:
"It is completely insane."
And counter to your "best known to no one" I've said before that I suspect system latency significantly contributes to FSDb's need to apply excess brake since the system makes decisions on data that could be upwards of 500ms to 1 sec earlier given data processing pipelines, filtering, and statistical decision making. This is a long known phenomenon in sensor targeting, ADAS etc. I don't buy the team's party line of 10ms. Making decisions based on 500ms to 1 second earlier data results in higher g-force and jerk applications when real time braking lead vehicles are iteratively closer and closer in real-time. It doesn't matter as much when lead vehicles are accelerating.
It's sad the team has put a low priority on improving vehicle dynamics and comfort when comfort is one of the 3 system goals. Gawd knows new NNs will introduce their own issues when attempting to control vehicle dynamics. Maybe the team didn't want to implement a quick, well performing heuristic design only to be disrupted by the inevitable need for NN data collection, training, revision and 'edge/corner' case chaos.
Last edited:
Similar threads
- Article
