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

FSD Beta 10.69

This site may earn commission on affiliate links.
I've noticed this too. However, back on 69.2.4 when it was showing the brake line, I noticed that regen is very aggressive, and there were many times I thought brakes were being used but it was all regen. When the car would blend in the brakes, there was definitely more inertial change. But 69.3.1 is not showing the brake line that I can see.
I think a difference is FSD has the ability to instantly use full regenerative braking. When you let off the accelerator it seems to ramp up more. I assume that’s to allow a smoother driving experience.
 
Here is the amazing video. You’ll see three instances of veering, each marked by guffawing/tittering.
New behavior. I have actually never seen this with any prior version.

Broken! Good enough for wider narrow release though it seems.

Amazing zero intervention drive though (except for navigation correction, and for a tiny accelerator application at the second stop sign). And yes, the first stop was a California stop.

Here I am trying to sneak by because you were positioned so far to the left.
1669843772775.png
 
Here I am trying to sneak by because you were positioned so far to the left.
View attachment 879924
Yes, it has no idea how to position itself in the road, that is clear.

Time for wide release! When is it coming?

Sadly it appears viral content is not as popular these days, so wide release will produce not even a ripple I suspect.

If the car could actually drive itself, it might end up making a substantially bigger splash. But as it is, it is just useless fluff.

Sorry, feeling quite pessimistic about the chances for success today.

It is so promising, but FSD can’t get the simple stuff right. I suppose we can hope and pray that they are leaving the simple stuff for last.
 
Yes, it has no idea how to position itself in the road, that is clear.

Time for wide release! When is it coming?

Sadly it appears viral content is not as popular these days, so wide release will produce not even a ripple I suspect.

If the car could actually drive itself, it might end up making a substantially bigger splash. But as it is, it is just useless fluff.

Sorry, feeling quite pessimistic about the chances for success today.

It is so promising, but FSD can’t get the simple stuff right. I suppose we can hope and pray that they are leaving the simple stuff for last.
Don’t feel bad - winter has arrived in MN which means FSD isn’t going to be an option for me half the time.
 
  • Like
Reactions: AlanSubie4Life
You’ll see three instances of veering
The first veering had a blue lane selection visualization indicating FSD Beta thought there was a fork, so it wanted to get into the "right lane" for an upcoming right turn. Later on similar streets, the visualization temporarily shows dashed white lines similarly reflecting the confusion of there being two driving lanes and wanting to get into the right lane.

temporary white lines.jpg
 
  • Informative
Reactions: AlanSubie4Life
The first veering had a blue lane selection visualization indicating FSD Beta thought there was a fork, so it wanted to get into the "right lane" for an upcoming right turn.
I don’t quite understand the point here. I saw that the blue path went to the right (the car always goes where it points). But why? Are you saying that maybe there was another lane detected but not certain enough to be visualized?

I see the markings on the second incident (I checked the first but had not bothered to examine this one in detail). It definitely seems like the lane logic is having some problems. It’s extremely evident that this road has a solid yellow down the middle of the broken road surface, wiped out in places, but evident enough. Note that when the veering took place, there was no such lane detection issue (visualized, anyway). There was a brief flash immediately before the veering - perhaps just enough confidence to visualize it?

It really struggles to figure out what is going on (see below in the instant before). I wonder when they will fix this stuff? Seems bad to not be able to determine these simple scenarios with 100% certainty. It’s very apparent there were no other lane lines, and even the one in the middle is basically a given even if it is worn off (obviously it is there; don’t even need to see it - the AI should do the same).

It’s hard to imagine easier scenarios.

6B50207A-4819-4F4F-9110-18785597A6DE.png


It did not like this! Sadly completely a non-issue. Large, obvious, 3mph slow down.

7C2EC47C-1E70-4DC7-8EEB-3ECE40C33027.jpeg
 
Last edited:
I saw that the blue path went to the right
I'm referring to the blue area not the blue path. It's hard to see in your case with the dark blue on dark background and goes away quickly as FSD Beta realizes there isn't actually a lane split. Here's a clearer view from Dirty Tesla where 10.69.3.1 also got confused but by some wet tire tracks:

confused lines.jpg


It even turned on the left turn signal to stay on the right side of the street? I wonder if it's more triggered by the driveway as a potential fork similar to your first veering with the driveway right next to a FedEx drop box?

In any case, this seems like it falls into the polish category where there could be a potential lane prediction sensitivity tradeoff resulting in FSD Beta not getting into a right lane when it needs to. This other end is also a polish issue as while annoying it missed a turn, it isn't really a safety issue.
 
  • Informative
Reactions: AlanSubie4Life
Thanks, much more clear, did not notice that.

In any case, this seems like it falls into the polish category where there could be a potential lane prediction sensitivity tradeoff resulting in FSD Beta not getting into a right lane when it needs to.
Maybe internally it is a polish issue from a technical standpoint, but end result is not just a polish issue. I just let it ride, since I was being covered by another driver behind me, but no normal driver would allow this.

These tradeoffs need to be crisp enough that there is essentially no tradeoff. The NNs need to be very very good indeed, with extremely low rates of false positives/false negatives.

For example, the tire tracks should never be interpreted as lane lines (they clearly are not).
 
  • Like
Reactions: pilotSteve
Thanks, much more clear, did not notice that.


Maybe internally it is a polish issue from a technical standpoint, but end result is not just a polish issue. I just let it ride, since I was being covered by another driver behind me, but no normal driver would allow this.

These tradeoffs need to be crisp enough that there is essentially no tradeoff. The NNs need to be very very good indeed, with extremely low rates of false positives/false negatives.

For example, the tire tracks should never be interpreted as lane lines (they clearly are not).
On the other hand if I didn’t know you were using FSD beta I would have assumed you were intoxicated and given you a wide berth. This is an under appreciated safety benefit of FSD.
 
Quick question for anyone who might know the answer.......I have FSDb 36.20 and would like to set my software to "advanced" in order to get 2022.40+. If I do this, do I out myself in jeopardy of not getting FSD V11 when its available? Thanks in advance.
Setting to Advanced has NOTHING to do with getting another update that is not Beta. You are on Beta and you stay on Beta unless you send an email asking to be removed from the Beta program.
 
I had an interesting observation yesterday morning - we had a snow storm and I was driving to work 'the old fashioned way.' The roads were covered in snow and not only were the lane markings not visible, for much of the drive there were not even tire tracks visible. I glanced down at the road visualizations a few times and not only did the computer do a pretty good job of estimating the edge of the road, it actually had lane markings in places. To be clear, there was nothing visible; I just drove based on where the edge of the road was, likely straddling the center line half the time, then slowing down and moving closer to the edge when an oncoming car approached.

Based on this it appears that FSD must use some map information combined with the cameras to generate the visualizations.
 
I had an interesting observation yesterday morning - we had a snow storm and I was driving to work 'the old fashioned way.' The roads were covered in snow and not only were the lane markings not visible, for much of the drive there were not even tire tracks visible. I glanced down at the road visualizations a few times and not only did the computer do a pretty good job of estimating the edge of the road, it actually had lane markings in places. To be clear, there was nothing visible; I just drove based on where the edge of the road was, likely straddling the center line half the time, then slowing down and moving closer to the edge when an oncoming car approached.

Based on this it appears that FSD must use some map information combined with the cameras to generate the visualizations.
The cameras also see just beyond human visual spectrum, and can compensate for glare. There have been many times, driving into the sun, that my human eyes lost the lanes - they were just gone, lost in a sea of glare and mirage reflections, but the car saw them perfectly. I'm not saying it saw through the snow, but if the snow was thin enough, they may have picked up the markings through the glare that we can't see.