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:
But it isn't just an issue on narrow unmarked roads. Yesterday I was driving in a different area than normal, and things were going fine, then the road I was on, which had very clear lanes lines, narrowed by a foot, or two, and suddenly the car was a scaredy cat, and slammed on the brakes as every car from the opposing direction got near, even though they were clearly fully in their own lane.

So it is interesting how just a foot or two totally changes how FSDb works.
My car did this on 11.3.4, but it seems fixed on 11.4.1 because that's how 50% of the roads I drive in are.
 
I feel like people are pretty aggressive in this forum with each other. I find this odd given that we're a relatively small group of people with very similar interests on a forum focused on that shared interest.
While we share interests here, we range from totally optimistic and positive fans to completely pessimistic and negative haters, so naturally there are arguments and sometimes aggressive jabs. For me, witnessing the amount of time naysayers devote to persistent negativity is a fascinating study in human behavior. Many of us ignore the worst of naysayers, but EVNow seems to be a glutton for punishment :)
 
It wants me to update to 2023.12.5.3
Anybody have more insights into what's in either 2023.12.5.3 or 2023.12.5.4? The former started rolling out the day after FSD Beta 11.4.1 and curiously only went to vehicles in US and Canada according to TeslaFi. The latter started rolling out this past weekend and just doubled pending from 4 to 8 vehicles presumably also only for US/Canada vehicles.

FSD Beta has only gone to vehicles in US/Canada, so could there be some shadow testing integration of FSD Beta 11.x into these 2023.12.5.x versions in preparation for a wide release on the main production software branch? Similarly the newer release could also be in preparation for 11.4.2, so Tesla might be doing some fleet testing to make sure vehicles with FSD Capability including those from 2023 are compatible?
 
Last edited:
While we share interests here, we range from totally optimistic and positive fans to completely pessimistic and negative haters, so naturally there are arguments and sometimes aggressive jabs. For me, witnessing the amount of time naysayers devote to persistent negativity is a fascinating study in human behavior. Many of us ignore the worst of naysayers, but EVNow seems to be a glutton for punishment :)
Totally. Most octogenarians I know are extremely passive.
 
So I've been on 11.3.6 for however long we all have (>1 month I'm pretty sure), but last week it changed it's lane selection behavior at two locations on my daily commute. Prior to last week, 11.3.6 chose the wrong lane 100% of the time. Since early last week (I think it was Monday's commute), it chooses the correct lane 100% of the time at these locations. I haven't received a map update, based on the version number on the software page.

One location is where a single lane splits into two, and the right lane only lasts for 0.5 miles before becoming a turn-only lane, and my commute requires going straight. So the best answer is to choose the left lane.
- Before last Monday, 11.3.6 chose the right lane 100% of the time, and if I disengaged, forced it into the left lane, and re-engaged, it would immediately try to change lanes to the right lane.
- Since last Monday, 11.3.6 has correctly chosen the left lane every single time, and stayed in it.

The other is at an intersection, where a single lane road enters the intersection, but it's two lanes exiting the intersection. The entry lane clearly lines up perfectly with the lane on the left, and almost all versions before v11 would go straight into that left lane. The left lane is doubly correct, because the right lane becomes a turn-only lane roughly 0.3 miles later, and my commute has me going straight there.
- Before last Monday, 11.3.6 would plan and execute a path swerving to the right in the middle of the intersection to enter the right lane (without using a turn signal)
- Since last Monday, 11.3.6 correctly goes straight into the left lane 100% of the time

I know I've felt like other versions have changed a bit over time, but I've always believed it was more likely just my perception of the behavior that was changing, or seeing random events go one way vs. the other. This time it was a very clear split with very measurable behavior, and I've got enough data points now (maybe a couple of dozen drives through these points before last Monday, and 6 since), I'm convinced it actually is acting differently as far as lane selection, even without an FSD update, and without a map update (at least the map with the version number on the software page, maybe there are different maps for lane data somehow).
 
So I've been on 11.3.6 for however long we all have (>1 month I'm pretty sure), but last week it changed it's lane selection behavior at two locations on my daily commute. Prior to last week, 11.3.6 chose the wrong lane 100% of the time. Since early last week (I think it was Monday's commute), it chooses the correct lane 100% of the time at these locations. I haven't received a map update, based on the version number on the software page.

One location is where a single lane splits into two, and the right lane only lasts for 0.5 miles before becoming a turn-only lane, and my commute requires going straight. So the best answer is to choose the left lane.
- Before last Monday, 11.3.6 chose the right lane 100% of the time, and if I disengaged, forced it into the left lane, and re-engaged, it would immediately try to change lanes to the right lane.
- Since last Monday, 11.3.6 has correctly chosen the left lane every single time, and stayed in it.

The other is at an intersection, where a single lane road enters the intersection, but it's two lanes exiting the intersection. The entry lane clearly lines up perfectly with the lane on the left, and almost all versions before v11 would go straight into that left lane. The left lane is doubly correct, because the right lane becomes a turn-only lane roughly 0.3 miles later, and my commute has me going straight there.
- Before last Monday, 11.3.6 would plan and execute a path swerving to the right in the middle of the intersection to enter the right lane (without using a turn signal)
- Since last Monday, 11.3.6 correctly goes straight into the left lane 100% of the time

I know I've felt like other versions have changed a bit over time, but I've always believed it was more likely just my perception of the behavior that was changing, or seeing random events go one way vs. the other. This time it was a very clear split with very measurable behavior, and I've got enough data points now (maybe a couple of dozen drives through these points before last Monday, and 6 since), I'm convinced it actually is acting differently as far as lane selection, even without an FSD update, and without a map update (at least the map with the version number on the software page, maybe there are different maps for lane data somehow).
It’s morphing and improving in the background. AI! Also, I have noticed the same thing things are rapidly improving for some reason. It still makes mistakes, but much less than when I first got the update.
 
  • Informative
Reactions: old pilot
More like keyboard warrior’s. Or sanctimonious Susie’s. I prefer the laughing Larry’s or the reality ricks myself. the delightful docs are fun too as well as the jabbin jebs and raucous rams. Couldn’t think of anything for @FSDtester#1

ok. I’m done
Your mom is checking in.

I will tell her you said hello!