Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
This site may earn commission on affiliate links.
The next big milestone for FSD is 11. It is a significant upgrade and fundamental changes to several parts of the FSD stack including totally new way to train the perception NN.

From AI day and Lex Fridman interview we have a good sense of what might be included.

- Object permanence both temporal and spatial
- Moving from “bag of points” to objects in NN
- Creating a 3D vector representation of the environment all in NN
- Planner optimization using NN / Monte Carlo Tree Search (MCTS)
- Change from processed images to “photon count” / raw image
- Change from single image perception to surround video
- Merging of city, highway and parking lot stacks a.k.a. Single Stack

Lex Fridman Interview of Elon. Starting with FSD related topics.


Here is a detailed explanation of Beta 11 in "layman's language" by James Douma, interview done after Lex Podcast.


Here is the AI Day explanation by in 4 parts.


screenshot-teslamotorsclub.com-2022.01.26-21_30_17.png


Here is a useful blog post asking a few questions to Tesla about AI day. The useful part comes in comparison of Tesla's methods with Waymo and others (detailed papers linked).

 
Last edited:
For reference, make sure you take videos of stopping manually, so everyone can see the smooth regen bar you're referring to.
Did a quick video. Small sample size, but I definitely did better than FSD! When manual driving, I had to tangle with a poorly-driven Model Y in front which led to a frustrating (but small - the scale is somewhat nonlinear I think?) regen excursion under manual control - but it was a single small pulse - not multiple pulses for a single stop, which is FSD's specialty, as you can see (an example here of a pulsed stop as well as a jam-on-the-brakes stop).


I could take these video samples every time I drive with similar results. This wasn't a cherry-picked video - I have others, all of them equally boring. There are worse screw ups, too. FSD doesn't do awful here - it just jammed on the brakes once, largely because it waited a second or two before slowing down for brake lights. But obviously any passenger would be saying “WTF?” due to this and other behaviors.

I would have disengaged multiple times to avoid the failure behavior, in normal circumstances.

This will be a waste of 8 minutes of your life. I don't recommend watching. At least do yourself a favor and set it to 2x speed so you only waste 4 minutes. Alternatively, I added some timestamps for the “important” stuff.

I don't plan to post any more video at this time. I rest my case.

The best direct comparison of FSD to human is 1:45 vs. 6:40. They are similar approaches to stopped traffic with no lead car. A stark difference in the regen bar and more importantly, anticipation. (This was not a horrible stop for FSD however - can be much worse as is seen elsewhere in the video.)

 
Last edited:
Did a quick video. Small sample size, but I definitely did better than FSD! When manual driving, I had to tangle with a poorly-driven Model Y in front which led to a frustrating (but small - the scale is somewhat nonlinear I think?) regen excursion under manual control - but it was a single small pulse - not multiple pulses for a single stop, which is FSD's specialty, as you can see (an example here of a pulsed stop as well as a jam-on-the-brakes stop).


I could take these video samples every time I drive with similar results. This wasn't a cherry-picked video - I have others, all of them equally boring. There are worse screw ups, too. FSD doesn't do awful here - it just jammed on the brakes once, largely because it waited a second or two before slowing down for brake lights. But obviously any passenger would be saying “WTF?” due to this and other behaviors.

I would have disengaged multiple times to avoid the failure behavior, in normal circumstances.

This will be a waste of 8 minutes of your life. I don't recommend watching. At least do yourself a favor and set it to 2x speed so you only waste 4 minutes. Alternatively, I added some timestamps for the “important” stuff.

I don't plan to post any more video at this time. I rest my case.

The best direct comparison of FSD to human is 1:45 vs. 6:40. They are similar approaches to stopped traffic with no lead car. A stark difference in the regen bar and more importantly, anticipation. (This was not a horrible stop for FSD however - can be much worse as is seen elsewhere in the video.)

It's sad after all these years they still haven't refined station keeping (and too many other features) to the level of a human driver. Admittedly it was much worse but there's still a long way to go. We can probably all agree there are many possible reasons like system latency, insufficient processing of visual distances, poor estimates of object kinematics, inability to capture human-like anticipation/context, poor design, less concern for driver comfort/confidence, bugs, ...
 
Admittedly it was much worse but there's still a long way to go
It definitely is a lot better than it used to be. The weird thing is that a decent percentage of the time it does really quite well with stopping now (sometimes I am sure it is going to jam on the brakes but then it does a perfect regen-only stop - happened just this morning, and on a downhill!). Maybe 1/3 to 1/2 the time it is ok? Yet the other stops are debacles of jerky pulsing stopping. It’s weird.
 
Did a quick video. Small sample size, but I definitely did better than FSD! When manual driving, I had to tangle with a poorly-driven Model Y in front which led to a frustrating (but small - the scale is somewhat nonlinear I think?) regen excursion under manual control - but it was a single small pulse - not multiple pulses for a single stop, which is FSD's specialty, as you can see (an example here of a pulsed stop as well as a jam-on-the-brakes stop).


I could take these video samples every time I drive with similar results. This wasn't a cherry-picked video - I have others, all of them equally boring. There are worse screw ups, too. FSD doesn't do awful here - it just jammed on the brakes once, largely because it waited a second or two before slowing down for brake lights. But obviously any passenger would be saying “WTF?” due to this and other behaviors.

I would have disengaged multiple times to avoid the failure behavior, in normal circumstances.

This will be a waste of 8 minutes of your life. I don't recommend watching. At least do yourself a favor and set it to 2x speed so you only waste 4 minutes. Alternatively, I added some timestamps for the “important” stuff.

I don't plan to post any more video at this time. I rest my case.

The best direct comparison of FSD to human is 1:45 vs. 6:40. They are similar approaches to stopped traffic with no lead car. A stark difference in the regen bar and more importantly, anticipation. (This was not a horrible stop for FSD however - can be much worse as is seen elsewhere in the video.)

I must have watched a completely different video? Looked perfectly fine to me. Car in front of you hammered brakes going from 45 to 20. Seems logical it would have the same reaction? If the car ended up within 2 feet of that car sure, but it maintained a safe zone which seems proper to me.
 
I must have watched a completely different video? Looked perfectly fine to me
Just watch the regen bar. Tells the story of manual control vs. FSD. Incontrovertibly fortunately.
Car in front of you hammered brakes going from 45 to 20. Seems logical it would have the same reaction? I

Did you notice the delay from brake lights to when FSD applied brakes? It was well over a second; you can measure it. 3:19 to 3:21 not counting partial seconds.

That’s the reason for the extreme braking (which was still not needed as there was plenty of room).

Traditionally, one immediately stops applying power when the car in front brakes. As I said, I would just have disengaged multiple times if I were not recording. Instant disengagement with stalk here when power continued to be applied.
 
Last edited:
I find myself curious how many OG 10.3 era FSDb folks are now on 2023.44.30.8?
My guess is most, if not currently, they will be soon. (I am.)

TeslaFi shows 10,780 vehicles not on it yet, 9,113 of those are on 2023.44.x, with 3,277 of those pending a 2023.44.30.8 install. So it seems like they are pushing this update to pretty much the entire fleet.
 
Think you mean 10.2 since 10.3 was a f'n DISASTER and only lasted about 4 days.:oops: I'm a 10.2er with 23.44.30.8. Expecting 24.2.69 in a couple of days with FSD 12.420. 🤣
I may not have been there for 10.2, but I was hooked seeing the videos and knew I had to have it. Just finished installing 23.44.30.8 and about to give it a try. With all now being loaded into this version, maybe this is the prelude to Version 12.
 
I must have watched a completely different video? Looked perfectly fine to me. Car in front of you hammered brakes going from 45 to 20. Seems logical it would have the same reaction? If the car ended up within 2 feet of that car sure, but it maintained a safe zone which seems proper to me.
It's difficult from the keyboard but not impossible. Just a little attention to detail in the video as our senses aren't confirming all that herky jerkiness.
 
Last edited:
Couple days back had to drive nearly an hour on smaller unfamiliar roads, at night and raining with blinding glare. I'd have been a nervous wreck if I didn't have FSD.

FSD can just figure out the lines much better than I can - and its not just my "old" eyes with glasses. I was asking my 10 yr old daughter - she couldn't see the lines most of the time too.

So much for calling it "j" ;)
 
Just watch the regen bar. Tells the story of manual control vs. FSD. Incontrovertibly fortunately.


Did you notice the delay from brake lights to when FSD applied brakes? It was well over a second; you can measure it. 3:19 to 3:21 not counting partial seconds.

That’s the reason for the extreme braking (which was still not needed as there was plenty of room).

Traditionally, one immediately stops applying power when the car in front brakes. As I said, I would just have disengaged multiple times if I were not recording. Instant disengagement with stalk here when power continued to be applied.
Yep. That delay tends to put brake control out of sequence with traffic flow which then requires even stiffer braking input. And as you mentioned, it is all so unnecessary as the FSD driver can see the red light well before the lead cars start slowing and later applying brakes. It's like watching a slow train wreck at each traffic light.
 
  • Love
Reactions: beachmiles
Did a quick video. Small sample size, but I definitely did better than FSD! When manual driving, I had to tangle with a poorly-driven Model Y in front which led to a frustrating (but small - the scale is somewhat nonlinear I think?) regen excursion under manual control - but it was a single small pulse - not multiple pulses for a single stop, which is FSD's specialty, as you can see (an example here of a pulsed stop as well as a jam-on-the-brakes stop).


I could take these video samples every time I drive with similar results. This wasn't a cherry-picked video - I have others, all of them equally boring. There are worse screw ups, too. FSD doesn't do awful here - it just jammed on the brakes once, largely because it waited a second or two before slowing down for brake lights. But obviously any passenger would be saying “WTF?” due to this and other behaviors.

I would have disengaged multiple times to avoid the failure behavior, in normal circumstances.

This will be a waste of 8 minutes of your life. I don't recommend watching. At least do yourself a favor and set it to 2x speed so you only waste 4 minutes. Alternatively, I added some timestamps for the “important” stuff.

I don't plan to post any more video at this time. I rest my case.

The best direct comparison of FSD to human is 1:45 vs. 6:40. They are similar approaches to stopped traffic with no lead car. A stark difference in the regen bar and more importantly, anticipation. (This was not a horrible stop for FSD however - can be much worse as is seen elsewhere in the video.)

You are too picky. I drive on the same road like you do but don't notice much. The most annoying break happens when some car on the left or right lane comes close to my car - it looks like FSD wants to yield the lane to those cars - bad.

Let Tesla fix my right turns first. Please, please. 🙏🙏 :)
 
I don't remember the firmware version when I bought my S90D on December 21, 2016 but I know I was on V10. I now have a 2023 S LR and I am on 2023.44.30.8.
Unrelated. You are referring to the UI version number and has nothing to do with FSD Beta. The current UI version is 11.1.

Just for reference FSD Beta 10.2 was released to Early Access in October 2021.
 
Last edited:
I don't remember the firmware version when I bought my S90D on December 21, 2016 but I know I was on V10. I now have a 2023 S LR and I am on 2023.44.30.8.

Unrelated. You are referring to the UI version number and has nothing to do with FSD Beta. The current UI version is 11.1.

Just for reference FSD Beta 10.2 was released to Early Access in October 2021.
FSD version history wikipost started by @EVNow

 
It's sad after all these years they still haven't refined station keeping (and too many other features) to the level of a human driver. Admittedly it was much worse but there's still a long way to go. We can probably all agree there are many possible reasons like system latency, insufficient processing of visual distances, poor estimates of object kinematics, inability to capture human-like anticipation/context, poor design, less concern for driver comfort/confidence, bugs, ...
1705012637976.png