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:
11.4.2 is bad. Really really bad. Its lane selection is the worst of any FSD beta release ever. On a simple 4 mile drive today in the suburbs I had 7 disengagements due to it selecting the right turning lane at intersections where we want to go straight. Weirdly, when it does this there is no “changing lanes to follow route” message on the screen. I have seen this behavior at some intersections before, but now it is doing it at even more intersections. Seriously?! I know it’s a hard problem but we are 3 years in now with FSD beta and it’s the worst it’s ever been. Is this ai team really all that great? I know FSD is a significantly harder problem, but there are other a.i. products out there that are actually useful now (think chatGPT.) Again, I know that human communication is significantly easier, but I feel like it is impossible to trust the FSD team at this point seeing as how we have made little-to-no progress really over the past few years regarding lane selection. Just something to think about.

Good questions and I imagine many are wondering the same thing. I don't think software churning is ever a successful strategy as issues tend to propagate and cause/effect becomes blurred. They seem to be taking full advantage of the FSD sucker base as the driver is ultimately responsible. Gawd knows how long they'll be allowed to get away with it.

Regarding lane selection I've noticed 11.4.2 almost consistently makes poor decisions whenever other vehicles are involved. Traffic can almost be non-existent and 11.4.2 will still plant you into a lane of an adjacent lane slowing vehicle or an adjacent lane passing vehicle. I can't even say it works well when no other vehicles are involved as 11.4.2 will make lane changes in intersections, solid lines, etc.
 
Last edited:
  • Like
Reactions: FSDtester#1
Another thing to add. Elon keeps talking about how much progress we have made. He even recently said (again) that we will achieve FSD safer than a human by the end of the year. How in the actual heck can he say this? He has access to the disengagement numbers and can see that there is no downward trend in disengagements at all. None. Seriously why.
 
He has access to the disengagement numbers and can see that there is no downward trend in disengagements at all. None. Seriously why.

Err, are you referring to the voluntarily reported disengagement data? There's literally only 155 people feeding into that tracker. Out of how many on FSD Beta now? Over 400k?

Elon has disengagement rates for the entire fleet, which are not shared publicly. I don't know if that data would show a downtrend, but the 155 people voluntarily reporting their data are neither representative nor a statistically significant sample.
 
  • Like
Reactions: GSP and STUtoday
Err, are you referring to the voluntarily reported disengagement data? There's literally only 155 people feeding into that tracker. Out of how many on FSD Beta now? Over 400k?

Is there some reason to think a substantive % of those 155 people are experiencing vastly different disengagement rates than anyone else?

FWIW with a population of 400,000 and a sample size of 155, if you target a 95% confidence interval your margin of error is 7.87%
While that's not awesome, it's small enough that no change is probably not due to overly small sample size unless the real change is single digits small.


Elon has disengagement rates for the entire fleet, which are not shared publicly.

Probably because those also don't show significant progress :)
 
Last edited:
Is there some reason to think a substantive % of those 155 people are experiencing vastly different disengagement rates than anyone else?

It's possible that people who live in areas where FSD Beta performs poorly are more likely to find and self-select into a tracker after Googling for solutions.

But that being said, I think there are actually some pretty clear trends over the last year. The percentage of drives with no critical disengagements has been trending towards 100%, and city distance between disengagements took a dip about 6 months ago, but has been on the rise since:

Screenshot_20230608-221817.png


But as I said before, I don't think we can extrapolate much from these because of the small sample size.
 
Regardless of build the one consistent and fairly dangerous problem I have with FSD is pulling out in front of cars instead of waiting. Or pulling out without fast acceleration. In 99% of the cases the view either left or right is obstructed mostly from shrubbery that appears during spring.

Here is a video where FSD committed to pulling out and I had to brake immediately or I might not be writing this post. I also have the clip just before this clip with the last frames showing braking to avoid the truck hitting me. Scarry moment.
Am I the only one convinced that if I drove FSD without careful supervision I'd have been in multiple accidents since I got FSD in October 2021?
Today FSD started to pull out in from of tractor trailer going 35mph only 110 feet from my car. There is NO way the B-pillar camera can see well enough to make a safe decision at that intersection. I always have to creep slowly and lean forward as far as possible and even then it's a challenge to make the decision when to go. Difficult intersection all year round but much worse since foliage came out. I still consider the lack of forward cameras FSD's achilles heel.

 
Last edited:
I'm going out on a limb here and saying something is wrong with your car. Either a hardware issue (cameras messing up, GPS antenna getting interference and not accurate, or faulty MCU or FSD computer), or a software issue (corrupt firmware installation, bad camera calibration, or bad USB drive).

I totally buy the weird little issues people have, and odd lane changes, etc. But the fact that your car will slow down/stop at a green light, or will sit at a red light and then not go when it's green, or turning into an oncoming lane - something is way off.
All this happened after the most “recent” map update.
 
It's possible that people who live in areas where FSD Beta performs poorly are more likely to find and self-select into a tracker after Googling for solutions.

But that being said, I think there are actually some pretty clear trends over the last year. The percentage of drives with no critical disengagements has been trending towards 100%, and city distance between disengagements took a dip about 6 months ago, but has been on the rise since:

View attachment 945183

But as I said before, I don't think we can extrapolate much from these because of the small sample size.
Define “critical disengagement.” Is it one where the car would crash? It must be a lower threshold than that? For me personally, I consider a critical disengagement to be anything that significantly affects other drivers around me. Changing lanes constantly unnecessarily pissing people off around me: critical disengagement. No one wants to ride in a car that makes people angry at all. My “look like an idiot” tolerance is quite low.
 
Happy to come along for the ride. Head west on the West Seattle Bridge, all the way to the end, and take the exit onto 35th Ave SW. That road all the way to the stop sign at Manning is hopeless. It's easy for us to recreate when there's nobody around since one of us can just drive the other direction toward the car running FSD, and the FSD car will come to a complete stop. It's embarrassing.

@SeattleFSD
Hopefully you can do this with @SeattleFSD - otherwise I’ll try to come after the summer vacation starts for the kids.
 
@Dave EV do you not have a model s/x? You disliked without replying…
No, I don't.

I got tired of arguing with people who claim that FSD is perfect - that it never breaks any laws, and that creating hazardous situations with erratic behavior often enough that you feel the need to put "student driver" stickers on your car is an acceptable idea, when you should just disengage FSD, take over and stop acting like you're the only person on the road and inconveniencing other drivers doesn't matter.
 
When did you last park the car? COP only lasts for 12 hours, then it has to be reset. If that's what's happening, I can give you a way to reset automatically.

This is a huge issue for me because I don't have a garage, and Arizona burns up car Interiors. I need my computer to last long enough for FSD to be solved.
Same day. I triggered the AC, let it run until 72, turned it off. After 10 mins or so COP randomly turned on. Junk.
No, I don't.

I got tired of arguing with people who claim that FSD is basically perfect - it never breaks any laws, creating hazardous situations with erratic behavior often enough that you feel the need to put "student driver" stickers on your car is a good idea, when you should just take over and stop acting like you're the only person on the road.
Are you implying I’m one of the people who thinks FSDj is basically perfect? Because if you are, you’ll get a lot of laughing upvotes on your post.
 
Are you implying I’m one of the people who thinks FSDj is basically perfect? Because if you are, you’ll get a lot of laughing upvotes on your post.
I have no idea - I disagreed because you LOLd in reply to one of my posts and implied I had no idea what I was talking about because I might have a Model S or X.

Which doesn't matter, by the way.
 
I have no idea - I disagreed because you LOLd in reply to one of my posts and implied I had no idea what I was talking about because I might have a Model S or X.

Which doesn't matter, by the way.
I think FSDj (j for Junk) is outright garbage, and it’s been showing a trend of being garbage on more S/X vehicles than 3/Y. I guess you haven’t been in this thread long enough to see my ranting 😅😅.

My comment to you was assuming you probably have the flashship refresh S/X vehicle and have the same garbage experience as I do.

@WilliamG @jebinc @FSDtester#1 @Pdubs @kabin
 
I think FSDj (j for Junk) is outright garbage, and it’s been showing a trend of being garbage on more S/X vehicles than 3/Y. I guess you haven’t been in this thread long enough to see my ranting 😅😅.

My comment to you was assuming you probably have the flashship refresh S/X vehicle and have the same garbage experience as I do.
Fair enough!

No, I can't keep up with this thread, that would require monitoring this thread alone to be a full-time job. :)

FSDb has it's glimmers of brilliance, but then it does some really weird, weird stuff.

Edit: removed my thumbs down, btw, clearly misunderstood your post.
 
Another thing to add. Elon keeps talking about how much progress we have made. He even recently said (again) that we will achieve FSD safer than a human by the end of the year. How in the actual heck can he say this? He has access to the disengagement numbers and can see that there is no downward trend in disengagements at all. None. Seriously why.
Because Elon is extremely optimistic about everything. There is exactly 0% chance that FSD will be safer than a human by the end of the year. I think every single person in this thread knows that, too.
 
Define “critical disengagement.” Is it one where the car would crash? It must be a lower threshold than that? For me personally, I consider a critical disengagement to be anything that significantly affects other drivers around me. Changing lanes constantly unnecessarily pissing people off around me: critical disengagement. No one wants to ride in a car that makes people angry at all. My “look like an idiot” tolerance is quite low.

I don't know if the designers of the tracker ever defined it. And even if they did, every person filling out the tracker probably has their own definition.

My entire point was that in terms of FSD Beta performance data, we have our own experiences, some anecdotes posted here by people having a bad time, some anecdotes posted on YouTube by people having a good time, and a random smattering of self-selected people in a tracker. None of it forms a comprehensive view of the FSD Beta disengagement rate, so to claim there's been no decrease overall is baseless.
 
11.4.2 has coding to limit the car’s speed based on the road. I’ve come to the conclusion it’s worthless. Below is a picture of one of the trunk roads in our neighborhood. It’s about 4.5 cars wide - wide enough for 2 cars to comfortably pass each other even with a cars parked on both sides of the road and actually has a yellow stripe in some areas. THere’s a sidewalk on both sides and a bike path in some areas. The posted limit is 30, people usually drive between 30-35 MPH. 11.4.2 now limits the speed to 30 MPH.

The second picture is a road by our cabin. It is barely wide enough for 2 cars to pass without driving off the pavement. Most people drive about 20 mph. FSD thinks the speed limit is 55 and tries to drive that fast.

I rest my case.

1686314153652.png
IMG_2469.jpeg