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:
I think they also mentioned the parking lot stack coming in a month or two?
Were there any indications that FSD Beta 11 single stack potentially including parking lot stack will have new functionality such as the long-awaited (since 2016) "parking space finder" / "park seek mode?" Or it'll first drive much better in parking lots as part of the usual FSD Beta city streets driving behavior?

Or as early access, maybe we'll be testing some of Park Assist, Autopark, Summon, Smart Summon switching to Tesla Vision instead of relying on ultrasonic sensors?
 
No one know for sure. FSD has always lagged behind. I’ve seen many speculate that 69.3 may be on a later version but no real data on it.
After reviewing fleet activity on TeslaFi, I suspect (hope) that 10.69.3 will accompany a s/w upgrade to 2022-28. I am eligible for FSD and have been idled on 2022.28.2 while roughly half the Teslas on 2022.28 have been migrated to 2022.36.1 through 2022.36.4. If I'm right, FSD-eligible units are being "held" on 2022-28 for the FSD convergence. Fingers crossed.
 
I think the biggest challenge I’ve dealt with on NoA is how it struggles with lane lines on occasion. Certain highway splits will cause the car to bounce or suddenly swerve a bit which can be jarring for cars behind you. Other thing I noticed was when a car merged in front of me the Tesla could no longer see the lane dash lines and swerved a bit as well. Just some things which I wonder will improve with FSD
All of that could be solved with the technology that Tesla/Elon doesn't want to use or pay for: better maps. In particular, instrument where average human drivers go on each lane, and how fast, and make a crowd-sourced consensus target for these. Then use that as a desirable target (not a 'rail') in the driving policy.

The highway splits would be solved very simply: since instrumentation of future paths of cars which drove there is available, the car can know "cars going in the direction of my destination typically drove *here* in the lanes", because humans understand semantics of how highways were built and don't rely exclusively on visual lane lines.

Of course the car should be able to drive in areas where there are no maps, or the maps are wrong, but in the common case when the maps are mostly right, it would drive better. It would be trained by thousands of humans.
 
  • Like
Reactions: sleepydoc
All of that could be solved with the technology that Tesla/Elon doesn't want to use or pay for: better maps. In particular, instrument where average human drivers go on each lane, and how fast, and make a crowd-sourced consensus target for these. Then use that as a desirable target (not a 'rail') in the driving policy.

The highway splits would be solved very simply: since instrumentation of future paths of cars which drove there is available, the car can know "cars going in the direction of my destination typically drove *here* in the lanes", because humans understand semantics of how highways were built and don't rely exclusively on visual lane lines.

Of course the car should be able to drive in areas where there are no maps, or the maps are wrong, but in the common case when the maps are mostly right, it would drive better. It would be trained by thousands of humans.
Tesla barked up this tree years ago and ultimately abandoned it. Their crowd-sourced ADAS tiles were scraped off Tesla's servers by folks on this forum and they were storing all the path data in there.
 
Tesla barked up this tree years ago and ultimately abandoned it. Their crowd-sourced ADAS tiles were scraped off Tesla's servers by folks on this forum and they were storing all the path data in there.
That seems like they failed to do the ML fusion in the planner well enough. Maybe then it was too hard-coded and perception too immature, but now they're getting to a better state with perception, with planning the clearly inadequate part given people's reports of failures. Maybe then they were trying to get perception of stationary objects as well (which can be done better now with occupancy networks).

Path of where humans actually prefer to drive cannot be discerned by any perception stack ever, it can only be done by measuring humans. And I don't believe it can be sufficiently compressed into a general-purpose network (though that's a good start), specific local knowledge is better.

I personally experience lots of substandard AP path decisions which could be solved with semantic maps. Besides, measuring human pathfinding is also useful as supervised inputs to a high-ML vs rules-based planner.
 
So, after the earnings call, I have questions.

Elon said FSD will be available to everyone this year that paid for it. Does this include subscriptions?

Based on past promises, I'm thinking not, at least not for those on the latest firmware who are not yet on Beta, despite months of high 97+ SS on two cars.

Even if some do get it, based on one case I experienced, well marked stop sign was not recognized, not ready for regulator approval.
 
So, after the earnings call, I have questions.

Elon said FSD will be available to everyone this year that paid for it. Does this include subscriptions?

Based on past promises, I'm thinking not

Why would you think that? A subscription requires payment.

not ready for regulator approval.

No regulator approval of any kind is required for wide release of an L2 system- which is what the current FSD beta is.
 
  • Like
Reactions: pilotSteve
Why would you think that? A subscription requires payment.



No regulator approval of any kind is required for wide release of an L2 system- which is what the current FSD beta is.
Are you sure that they have the right hardware? I.e cameras.
I am on the subscription. will I get FSD just like those who paid full boat?

Elon is saying the FSD will actually be FSD, and only require regulatory approval to be full L5. You can listen to the earning and make your own opinion.

I'm thinking he forgot about all those poor bastards with hardware 2 on 2019 and prior cars.
 
  • Like
Reactions: pilotSteve
I am on the subscription. will I get FSD just like those who paid full boat?

Why wouldn't you?


Elon is saying the FSD will actually be FSD, and only require regulatory approval to be full L5. You can listen to the earning and make your own opinion.

No, he's absolutely and clearly not saying that

He's saying THIS years wide release (in NA) will be L2, just like it is now- but safe enough for wide release.

He said he thinks it'll be the one that needs regulatory approval before end of NEXT year.

Which is a thing he's said for many years and always been wrong about, and you don't even need approval for L5 in half a dozen states right now anyway so we can safely ignore it once again.



I'm thinking he forgot about all those poor bastards with hardware 2 on 2019 and prior cars.

Anyone who has FSD has HW3 (or gets a free upgrade if say a HW2 owner buys FSD tomorrow). All 3/Ys ever made have the same cameras the current ones have, nothing special about 2019.

Some older 2017/2018 S/X cars need camera upgrades- but they also get those free if they have FSD just like the computer upgrades.
 
I am on the subscription. will I get FSD just like those who paid full boat?

Elon is saying the FSD will actually be FSD, and only require regulatory approval to be full L5. You can listen to the earning and make your own opinion.

I'm thinking he forgot about all those poor bastards with hardware 2 on 2019 and prior cars.
Anyone with a current subscription can request access to the FSD Beta, and as long as they pass the safety score, or Tesla removes the safety score, and Tesla opens the Beta up to more people, you'll get in.

If you let your subscription lapse, FSD Beta will be removed from your car.
 
Why wouldn't you?




No, he's absolutely and clearly not saying that

He's saying THIS years wide release (in NA) will be L2, just like it is now- but safe enough for wide release.

He said he thinks it'll be the one that needs regulatory approval before end of NEXT year.

Which is a thing he's said for many years and always been wrong about, and you don't even need approval for L5 in half a dozen states right now anyway so we can safely ignore it once again.





Anyone who has FSD has HW3 (or gets a free upgrade if say a HW2 owner buys FSD tomorrow). All 3/Ys ever made have the same cameras the current ones have, nothing special about 2019.

Some older 2017/2018 S/X cars need camera upgrades- but they also get those free if they have FSD just like the computer upgrades.
I was referring to those older MS cars.
 
Anyone with a current subscription can request access to the FSD Beta, and as long as they pass the safety score, or Tesla removes the safety score, and Tesla opens the Beta up to more people, you'll get in.

If you let your subscription lapse, FSD Beta will be removed from your car.
I've had the subscription for a month, and high SS for several months, but was never clear on how to request beta access. The SA did not know either.
 
I've had the subscription for a month, and high SS for several months, but was never clear on how to request beta access. The SA did not know either.
Have you gone into to the Autopilot settings in your car and opted in to the FSD beta? (That is how you normally get the safety score, so I'm not sure how you got the safety score before you subscribed to FSD. Unless you have Tesla Insurance, and opted to get the SS that way.)
 
Have you gone into to the Autopilot settings in your car and opted in to the FSD beta? (That is how you normally get the safety score, so I'm not sure how you got the safety score before you subscribed to FSD. Unless you have Tesla Insurance, and opted to get the SS that way.)
I do have Tesla insurance.

I stopped my subscription when I learned beta was not being offered on the lastest firmware I just installed.

I will have to see if the .36 fixes are worth not having access to Beta for a while.
 
To think this tread is now almost a year old and V11.x is nowhere is sight. Except Elon is still testing it. Who would have thought that we were unlikely to even get to V11.x in 22 last year. Looks like the odds of V11.x making it this year are now kinda low. Sounds like "one stack to rule them all" is a LONG term challenge.
 
  • Like
Reactions: aerodyne
Sounds like "one stack to rule them all" is a LONG term challenge
From the most recent AI day and quarterly earnings call, it seems like the current top priority is wide release without safety score requirements. This seems to be based on their internal metric of safety critical interventions.

The next version 10.69.3 has at least improvements to high velocity cross traffic predictions, so that aspect isn't as relevant to interstate driving with controlled access ramps. Potentially there's occupancy network improvements for safer parking lot driving too, but this might actually delay FSD Beta 11 replacing highway stack if a more complex occupancy network needs to run at highway speeds.

So it sounds like single stack is lower priority because existing Navigate on Autopilot is already quite safe and capable, but if wide release happens in November, the team will probably focus on the next priority hopefully being highway single stack (if not superceded by some other time sensitive issue like ultrasonic sensor feature parity).

Although one benefit of this quarter is that shadow mode single stack should be able to validate with the fleet driving in more nighttime / rainy / snowy situations.
 
  • Like
Reactions: RabidYak