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:
My second tesla - model Y (no fsd on this car) without USS just got update to 2023.6.9 which has autopark support

Downloading and installing...

Very excited.. I missed parking assist the most in my tight garage parking...yay!

View attachment 920832

Interesting. I got 2023.6.8 on my Model Y without FSD today and it was nothing but some useless to me language support.

IMG_0104.png


I mean… I guess at least it’s on the 2023.X branch 🤣🤣
 
Nope never heard of it. Starting driving a long time ago which I guess explains it. I just merge in as gracefully as possible. It's interesting how many YouTube videos complain about FSD waiting too long till the end to merge.

The zipper merge isn't a regulation or requirement; it's the most courteous way to merge when two lanes become one. More so on local street where neither lane had a greater right-of-way than the other. People who tailgate to prevent a merge are considered the masshole by those around them because they are violating the 1:1 nature of a zipper merge.

A highway merge is a bit different, since the on-ramp lane usually has to yield to the highway lane. But in bumper-to-bumper traffic, typically the merging pattern evolves into a zipper. And ideally the on-ramp cars move as far forward as possible in the lane, as this reduces backup congestion (you're using up all available real estate the road provides you and giving more cars the opportunity to merge).

The biggest problem I see here with highway merges is when on-rampers don't have the sense to just wait for a tractor trailer to move forward. They all try to squeeze in next to it to get in front of it... until they can't, and then someone inevitably has to do the "hang back of shame" and go behind the tractor trailer. FSDb likely has this problem.
 
What’s the deal with this @STUtoday guy? Is he some sort of dislike bot? He trolls these threads and dislikes posts all the time, never see him like any posts, nor ever see him actually post anything.

Mojo is that you?!? Are you Elon’s dislike bot?
He’s never disliked anything I’ve posted. Maybe you’re just too radical for him. But then again, everything I post that’s remotely funny, gets moved to the other thread. Maybe it’s Krash?
 
Auto park has been around for years, before cars had 360deg cameras. AP1 Teslas had autopark. Old Lexus’ and Fords had self parking features before surround view cameras.
I don't think you need surround view per se, but you need front/rear cameras and AFAIK all autopark systems are mainly camera-based with the USS there for the fine tuning and short-range object detection while maneuvering.

AP1 used Mobileye's hardware and I wouldn't be surprised if most or perhaps all autopark systems back then were using implementations from Mobileye
 
Autopark systems across the industry are vision-based, hence why they require vehicles on either side or visible lines for the parking spot
It's hard to unpack this statement, but if it is suggesting that most autopark systems (i.e. automated parallel parking) across car manufacturers are vision-based or have traditionally been vision-based, then this is just factually incorrect. Autopark systems from car manufacturers have traditionally been, and most still are, USS based.
 
Zipper merges with many FSD cars on the road will be interesting as in Germany, the rules are to use up both lanes and then merge at the very end, while in the US, it is perceived to be rude to use the mostly empty disappearing lane to pass up the single line of cars queued up in the lane that doesn't disappear well in advance of the merge point. Should be simple to program, but it will required two different behaviors.
 
  • Like
Reactions: Pdubs
It's hard to unpack this statement, but if it is suggesting that most autopark systems (i.e. automated parallel parking) across car manufacturers are vision-based or have traditionally been vision-based, then this is just factually incorrect. Autopark systems from car manufacturers have traditionally been, and most still are, USS based.
Do you have one of these traditional vehicles? I'd be curious how the Autopark would react with the cameras covered.

Activating the Autopark in my non-Tesla is done by holding down a Camera button.
 
Zipper merge works better in heavy slow moving traffic where there's more time to make decisions. Of course as speeds increase there's less time to negotiate a safe zipper merge before the merge lane ends. After all, FSD should primarily be about safety. FSD frequently struggles when pressed for time so it's a bad mix.

I suspect V11.3.2 didn't intend to implement a zipper merge. Instead an arguably easy implementation was to follow the right lane edge until it disappeared. Check the box and just ship it baby. What could possibly go wrong.
 
  • Funny
Reactions: aronth5
For fun it would be interesting to compare various large language model AI chatbot responses to the question "why did Tesla stop the FSD v11.3.2 wide release?"

Here's Bard's response:

FSD v11.3.2 wide release was halted because some Tesla owners reported that it was causing their cars to make dangerous mistakes. For example, some cars were reportedly turning left into oncoming traffic or stopping in the middle of the road. Tesla is investigating these reports and will release the update again when it is safe to do so.

This is not the first time that Tesla has halted an FSD update due to safety concerns. In January 2022, Tesla halted an FSD update after some cars were reportedly making mistakes that could cause accidents. Tesla is committed to safety and will not release an update that could put drivers at risk.