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:
It can change lanes, I don’t think it will do other turns but I disengage before I turn so if it can, I haven’t seen it.
I suspect that too. So, it's basically Auto Steer feature.
It will be great if Tesla has option: FSD with manual routing. This will eliminate random lane changing and late lane changing to make turns or to enter/exit freeways.
 
Last edited:
I suspect that too. So, it's basically Auto Steer feature.
It will be great if Tesla has option: FSD with manual routing. This will eliminate random lane changing and late lane changing to make turns or to enter/exit freeways.
I would love that. I can basically drive home from work in the same lane, excluding my driveway, but FSD loves changing lanes for no reason several times in the same places, even with minimal lane changes on.
 
  • Like
Reactions: Genie and rlsd
FWIW, release notes for v11.4.6. Identical to v11.4.4/5, except for one additional bullet point:

- Introduced Automatic Emergency Braking on general obstacles detected by Occupancy Network.
I wonder if this is in response to Don O'... (If his fake kids/dogs are detected as "general obstacles" instead of a person/dog.)
 
It can change lanes, I don’t think it will do other turns but I disengage before I turn so if it can, I haven’t seen it.
If you come to a T intersection (can't go straight) it will make a turn, usually a right turn.

EDIT: Unless something has changed in the last year or so since that was the last time I engaged without a distention.
 
Last edited:
  • Informative
  • Like
Reactions: APotatoGod and rlsd
This past week, I drove for four or so hours on a non-FSD car, and for all the complaints I have with FSD, I would still rather have it in its 11.4.4 configuration than not have it at all. I can relax with FSD on the freeway system, and it really makes driving more comfortable for me.

Joe
That's what I worry for my far away trip next month: I have to drive a Chrysler Minivan for couple days.
 
I was surprised at how quickly I adapted to using FSD, and how much I miss it when I am driving a vehicle without FSD. The Cadillac I drove had adaptive cruise control, but it cannot steer nor change lanes to move around slower vehicles. I constantly found myself checking the speed limit and then manually driving around slower vehicles. It isn’t as if any of this driving is difficult, but it is SO NICE to be able to relax a bit, especially on long drives. On city streets, FSD seems to enjoy trying to kill me. On the interstate; oh my gosh is it nice.

Joe
 
I know there are imperfections, less so now that I use FSDb without a navigation destination, but sometimes it does something that is so amazing it renews my excitement. Driving along my normal route this morning, the car suddenly started to brake for no visible reason, just as I was about to cuss it out, a deer darted across the road, even showing up on the visualization (legs don't move so it's kind of funny) I totally did not see it at least 3 full seconds before the car did. Crazy cool!
 
I know there are imperfections, less so now that I use FSDb without a navigation destination, but sometimes it does something that is so amazing it renews my excitement. Driving along my normal route this morning, the car suddenly started to brake for no visible reason, just as I was about to cuss it out, a deer darted across the road, even showing up on the visualization (legs don't move so it's kind of funny) I totally did not see it at least 3 full seconds before the car did. Crazy cool!
I wish my Amazon Ring camera could tell the difference between a tiny spider and a big burglar. It keeps giving false alarms every night. 😁
When will Jeff Bezos start using AI vision?
 
  • Funny
Reactions: APotatoGod
I wish my Amazon Ring camera could tell the difference between a tiny spider and a big burglar. It keeps giving false alarms every night. 😁
When will Jeff Bezos start using AI vision?
OMG I thought I was the only one with this issue lol. Clear the web then a few days later the chime is going off every 10 minutes in the middle of the night while a spider climbs Up and Down in front of the lense lol.
 
OMG I thought I was the only one with this issue lol. Clear the web then a few days later the chime is going off every 10 minutes in the middle of the night while a spider climbs Up and Down in front of the lense lol.
The infrared light on the Ring camera glows at night and attracts bugs. Spiders come to build webs to catch bugs. Amazon engineers don't understand that. LIDAR is bad. 😂
 
If 10s or 100s of billions of miles driven is necessary to train the AI / NN to get better than a human, then why is the barrier to beta test FSD so high (i.e., $15 k). It's robotaxi value is zero today. When the day comes, and someone starts using it as a robotaxi, charge them the $15,000 then, not now. Sorry if this is off topic, but I think it is at worst, tangential.
 
If 10s or 100s of billions of miles driven is necessary to train the AI / NN to get better than a human, then why is the barrier to beta test FSD so high (i.e., $15 k). It's robotaxi value is zero today. When the day comes, and someone starts using it as a robotaxi, charge them the $15,000 then, not now. Sorry if this is off topic, but I think it is at worst, tangential.
This was pretty much my thinking that led to this thread I started yesterday suggesting different pricing depending on the intended use case.
 
  • Like
Reactions: pilotSteve
This was pretty much my thinking that led to this thread I started yesterday suggesting different pricing depending on the intended use case.
Actually, my read on this is somewhat different.

Take two developments. One has houses plotched down and ready to sell.

The other has Nothing But Plans. Buy the house before built; the houses, for this example, would be the same.

Guess which one is cheaper? Not much need to tear one's hair out: Houses bought before building are lots cheaper. (Which covers the place the SO and I are living in, now. Another attractant is that one can make cheapie changes to a house before it's built: Different places for an outlet; better house insulation; better SEER on the air conditioning system, and so on. Yes, it costs a bit more to make those changes, but the amounts are far less than to make those changes after the fact.)

FSD is like that. Early adopters got it for $2k; I picked up my option at $6k. Tesla has been clear: The closer FSD is to being complete, the more it'll cost.

Admittedly, they've been slower than expected getting the full robo-taxi out the door. To which I state: What do you want for a research project?
 
  • Like
Reactions: FSDtester#1
Just saw there's a recall for a good portion of 2023 HW4 vehicles over a camera pitch issue, wonder what impact that has on performance or even the training at this point, if any?
A good portion? They think it is only a problem for ~1,000 vehicles. That is like about a single day's production worth. (But the impacted vehicles were built between January 17th and July 11th, so it seems like a minor portion of HW4 vehicles to me.)

As far as performance, this is what the recall says:

The forward-facing camera may be misaligned, causing some of the active safety features such as emergency braking, forward collision warning, and lane assist to become unavailable without alerting the driver.

But yet they had 83 warranty claims to fix it. How did those 83 drivers know that something was wrong? (Maybe the cameras weren't calibrating and AP/TACC wouldn't work either?)
 
  • Informative
Reactions: FSDtester#1
A good portion? They think it is only a problem for ~1,000 vehicles. That is like about a single day's production worth. (But the impacted vehicles were built between January 17th and July 11th, so it seems like a minor portion of HW4 vehicles.
Not sure. I know they delayed a lot of S and X owners who ordered in 2022 and gave them the option to take an inventory vehicle as it seems S and X built during that timeframe were hardware 4. I remember the threads being on fire with VIN number ranges trying to determine where the first came off the line and if people should refuse delivery to get an HW4 vehicle. The way I read it was a large portion of Hw4 but that could mean that there weren’t that many HW4 rolling off the line to begin with. I suppose tube will tell. Now that you mention the 83 people I also wonder what tipped them off, an error code, wonky performance, etc
 
If 10s or 100s of billions of miles driven is necessary to train the AI / NN to get better than a human, then why is the barrier to beta test FSD so high (i.e., $15 k). It's robotaxi value is zero today. When the day comes, and someone starts using it as a robotaxi, charge them the $15,000 then, not now. Sorry if this is off topic, but I think it is at worst, tangential.
Dang, autocorrect didn't fix my It's to Its.
Maybe their training engine can't handle 100s of billions of miles driven yet, and don't want more drivers/miles at the moment.