Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
  • Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
  • Tesla's Supercharger Team was recently laid off. We discuss what this means for the company on today's TMC Podcast streaming live at 1PM PDT. You can watch on X or on YouTube where you can participate in the live chat.
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:
and I’ve already had my car for a couple of years and will 100% trade it in for a newer model before the junk is worth anything. $10k to beta test some software, lol. Lesson learned the hard way.

I’m waiting for them to offer a transfer of FUSC and FSD, then maybe I’ll buy another car.

Soon they’ll be desperate enough, at least according to Elon; it sounds like he was saying they can’t sell cars anymore, according to the latest earnings call.

Even so, seems likely FSD will still be no good, even for the lifetime of the new vehicle. But such is life.
 
It would be an interesting alternate universe if Tesla had concentrated on mastering AP/FSD on highways first -- instead they've completely lost their lead in the practical side of self-driving in pursuit of the robo-taxi dream. Of course, maybe they wouldn't have received all the free advertisement they got from proclaiming their cars would be driving themselves soon-ish.

Even though I only use AP and FSD for fun it bothers me that my wife's ID.4 can drive itself on the highway better than the Tesla can (no phantom breaking, capacitive steering wheel, free auto lane changing, ability to set what side of the lane you want it to drive on, etc. . ). Sure there's a chance Tesla can improve -- I don't have much faith seeing how long some of the basic failings of AP have been around -- especially phantom braking (not a problem with the ID.4). Why can the ID.4 solve phantom braking (or at least be a 100 times better) while the company that's trying to have their cars drive themselves everywhere can't?

I mentioned this before, but I steered a friend away from getting a Tesla when he told me he placed a high value on driver assistance/advanced cruise control tech. I told him that Hyundai/Kia/Volkswagen had passed Tesla in this area -- especially the free versions.
 
Last edited:
A little OT but I find it kinda odd that Elon has gone full "gag order" on V12 since the demo. Usually he continues to ramp up the hyper hype about each mind blowing drive he has. At first I thought this is maybe because fast progress was being made. But it is very unlike Elon not to boast about advances for so long.
It seems FSD has been one aspirational pipe dream after another (pure vision, photons, mind blowing, single stack to rule them all).

For v12 E2E, Elon said something like 'didn't realize how easy it was,' 'no heuristics,' and 'less training data needed.'

Elon chooses to ignore his rule of thumb 'prototypes are easy, production is hard' when talking about FSD. I'd say it's like a crowdfunder but crowdfunders might be held to higher ethical standards.
 
Remember V12 needs a large GPU buildout of Nvidia and Dogo. Think dev team is perfecting the code and a second team is building out the compute infrastructure. To get v12 out in public, even as beta, it’s a massive effort. They are heads down on that effort.
Even more of a challenge, Elon says v12 won't be beta. Elon has used up all his FSD good will and then some. I'll only believe it after I drive it.
 
Remember V12 needs a large GPU buildout of Nvidia and Dogo. Think dev team is perfecting the code and a second team is building out the compute infrastructure. To get v12 out in public, even as beta, it’s a massive effort. They are heads down on that effort.
I know this and not what my question was. I'm wondering why Elon is no longer Xing/raving about V12. That is totally anti-Elon behavior. He usually gushes on and on and on...... about an upcoming version.
 
I know this and not what my question was. I'm wondering why Elon is no longer Xing/raving about V12. That is totally anti-Elon behavior. He usually gushes on and on and on...... about an upcoming version.
It could be its already solved level 5 autonomy and Elon is just waiting to make the announcement for Christmas 🎄 😉!
 
It could be its already solved level 5 autonomy and Elon is just waiting to make the announcement for Christmas 🎄 😉!
Maybe
@FSDtester#1, we are watching Elon evolve as a mid aged man:
Keeping the cat in the bag, v12 if ready, and waiting for the best moment when ready
Or
Mature enough now to listen to his lieutenants, waiting for it to be ready
Especially with what just happened to Cruise and with Dowd/Dr. Mary “Missy” Cummings wait view

This is one of the biggest moments for Tesla
Similar turning point to #1 vehicle in world Model Y
FSD moving from beta to prod, pre Robotaxi will be Teslas biggest achievement

Every release of v11 FSDb waiting for V12, is like Christmas morning
Fun having a vehicle so relevant to leading edge tech
 
Maybe
@FSDtester#1, we are watching Elon evolve as a mid aged man:
Keeping the cat in the bag, v12 if ready, and waiting for the best moment when ready
Or
Mature enough now to listen to his lieutenants, waiting for it to be ready
Especially with what just happened to Cruise and with Dowd/Dr. Mary “Missy” Cummings wait view

This is one of the biggest moments for Tesla
Similar turning point to #1 vehicle in world Model Y
FSD moving from beta to prod, pre Robotaxi will be Teslas biggest achievement

Every release of v11 FSDb waiting for V12, is like Christmas morning
Fun having a vehicle so relevant to leading edge tech
Not a single car on the road right now, manufactured by Tesla, will ever, ever, ever be a robotaxi. I can guarantee it as much as I can guarantee anything.
 
Why can the ID.4 solve phantom braking (or at least be a 100 times better) while the company that's trying to have their cars drive themselves everywhere can't?
It's easy to solve .. they just do what every other car maker seems to be doing .. tune it so the bias is against false positives. But of course this ALSO leads to a reduction in TRUE positives .. in other words, you ID.4 wont phantom brake, but it also runs the risk of NOT braking when it should.

I'll take the odd phantom braking event over having the car crash in an emergency any day.
 
I'll take the odd phantom braking event over having the car crash in an emergency any day.
And then there is the sort of thing where it is cruising by someone on the freeway with a blinker left on, then right as the car is alongside said vehicle, it taps the brakes!

Safety!

😂

This sort of stuff is unacceptable. It’s just super low quality, regardless of any efforts made to have a few false positives in the name of safety.

And I don’t see how v12 would fix this.
 
And then there is the sort of thing where it is cruising by someone on the freeway with a blinker left on, then right as the car is alongside said vehicle, it taps the brakes!

Safety!

😂

This sort of stuff is unacceptable. It’s just super low quality, regardless of any efforts made to have a few false positives in the name of safety.

And I don’t see how v12 would fix this.
I wondered if FSDb is already checking on and considering following vehicles before braking? I have not heard of rear end collisions of moving Teslas on FSD due to PB.
 
I wondered if FSDb is already checking on and considering following vehicles before braking? I have not heard of rear end collisions of moving Teslas on FSD due to PB.
Bit somewhere I heard they are running V12 in v11 vehicles in the background to see how it performs, meaning the v12 stack watches and then says it would….
Then they can compare to v11 and what happened
Not an exact science, but a way to run v12 and gather simulated results
Pretty cool
 
  • Funny
Reactions: jebinc
I'm still highly doubtful they send much data back from the fleet. The cost for that would be astounding from just a data perspective, especially when you get into big video files. Some telemetry sure.
If you doubt it, and if you have your FSD car connected to your Wi-Fi, check the logs of your router. It's particularly easy if you have one of the routers or mesh systems that have an app for reporting and control.

As I understand it, Tesla prefers to collect the data from cars when they're connected to Wi-Fi, typically at the owner's home while idle or charging. This is probably to control costs as you referenced; the owner's high-speed internet connection avoids cellular data charges to Tesla.​
The same is true regarding software update downloads to the car- the user's Wi-Fi connection is used if and when available, but Tesla can use the cellular Connection of the car if they decide the data transaction is sufficiently important or time sensitive (e.g. if they're specifically targeting certain training data through a data seeking "campaign", or if the car is involved in a detected accident, or if a software update is considered urgent due to a recall deadline or other critical bug fix).​

For the normal data transactions over Wi-Fi , you can easily prove to yourself that significant amounts of data are uploading. In my case, my eero app (for my mesh Wi-Fi network at home) shows 2.9 GB uploaded from the car just for October 29th, the first day of the new week - fairly normal in my experience. This despite the fact that I have taken only two modest and completely routine commutes over the past 24 to 36 hours. The download data size for the same date is 2.5 MB which I suppose is simply data traffic overhead, the data request commands from the Mothership, handshaking, error correction and so forth.