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:
They've played the cards they had for HW3

Hmm, that won't get anywhere close to 1x.
Still some cards to play for HW3 after all. This improvement I believe was unexpected by most.
"the ability for the vehicle to stop when backing up and by increasing the top speed it can operate in a new update."

Tesla makes Automatic Emergency Braking even safer in update
My only point was to show that Tesla has in fact played another card for HW3 when you stated all the cards have been played. Stating absolutes like you did is always a risky thing to do. Who knows what else may be added to leverage HW3. Maybe yes, maybe no but only time will tell.
 
News today of another dust storm with dozens of vehicles piled up on the freeway. Six killed. In 2021 seven were killed in UT. In 2022 six were killed in MT.

Just another reason to opt for radar sensors on vehicles.
Don't those vehicles have radar with AEB (most that have adaptive cruise control)? I think it's more important to get the logic correct. Remember that Cruise slammed into a bus and it has loads of LIDAR and RADAR all over it.
 
Not likely.

That is FSD Beta V11.4. Elon has already mentioned that V11.4.1 is looking promising, but we have no idea what actual version it is. It could be 2023.6.16, or it could be 2023.12.x...

And we don't even know if V11.4.1 will be released to the public...
 
All other parts of Interstate driving seems so much better on Beta than legacy AP stack but…….why is the Interstate logic so bad at anticipating needed strategy to exit. Waiting on the last mile is fine if you are on 3 lanes of moving Interstate. However in urban settings and needing to get over 6 or 8 lanes with heavy, slow, inconsistent speed traffic you can’t wait to the last mile.
The main thing lacking in FSD is traffic awareness. We drive differently in heavy traffic vs light traffic. FSD drives the same way. Given they have the info, they should do better.
 
Looks like TeslaBull is still on 11.4 (latest YouTube). So if 11.4.1 is so promising at least let the employees get it Elon. 🤔 🤣

Also would love to see if there is change of it being on 23.12.x. Maybe the TeslaBull (still an enigma that this employee seems to have Elon's blessing to post) will get it before the weekend and we can see the Notes.

 
My only point was to show that Tesla has in fact played another card for HW3 when you stated all the cards have been played. Stating absolutes like you did is always a risky thing to do. Who knows what else may be added to leverage HW3. Maybe yes, maybe no but only time will tell.
Maybe so but unfortunately there have been consistent FSD absolutes one can bank on. I wouldn't be surprised if 11.3.x was HW3's last hurrah with 11.4 and beyond destined for HW4. At some point they will break off for HW4 data differences and potential performance improvements.
 
Don't those vehicles have radar with AEB (most that have adaptive cruise control)? I think it's more important to get the logic correct. Remember that Cruise slammed into a bus and it has loads of LIDAR and RADAR all over it.
Yep. They should be able to sense slowing objects within the dust storm, slow and steer accordingly versus vision-only punting to the driver which is what contributes to these terrible accidents.

Autonomous systems will always have bugs even if implementing good logic. The trick is redundancy and gracefully failing in an optimally safe manner.
 
Yep. They should be able to sense slowing objects within the dust storm, slow and steer accordingly versus vision-only punting to the driver which is what contributes to these terrible accidents.

Autonomous systems will always have bugs even if implementing good logic. The trick is redundancy and gracefully failing in an optimally safe manner.
But they didn't - dozens of them crashed. AEB didn't even kick in and save them - a standard safety feature of most cars today. Where were the ultrasonics on these cars, or the forward-facing radar for adaptive cruise control?

And where was the common sense from the drivers? A massive dust storm with near zero visibility, and people are thinking "sure, I can drive in this..."
 
So first drive today, still junky. Once red hands come up no matter how hard I jerk off the yoke, they still flash. After previous strikes I no longer find the y intercept on the **** around and find out graph, so pretty much forced to disengage once that behavior appears to avoid getting a strike.
maybe you shouldn't be jerking off the yoke. It might not like that from you. Have you had a female try it?
 
Since traffic can be different in different locals as well as different times of the day, why not have an option in the software to set the distance we would like the car to move over?
Not sure that will solve the problem.

I want FSD to figure out what is the right time to move over based on traffic. Mind you, not just highways, but also city.

So many times, FSD will not change lane, even though it is indicating an upcoming lane change, when the lane is free (and no traffic in front before the turn comes). Or in my case today - it had to move over to the opposite lane to go around a cyclist. But the visibility to other lane is low because of a hill crest. We would wait to crest the hill before going around the cyclist - and yesterday it tried to go around even when a car was oncoming, squeezing between cyclist and the car. Most of us would wait for the oncoming car to pass before going around the cyclist. But ... if the traffic is heavy we would try to squeeze in between. So, knowing what to do depending on traffic is important.
 
Maybe so but unfortunately there have been consistent FSD absolutes one can bank on. I wouldn't be surprised if 11.3.x was HW3's last hurrah with 11.4 and beyond destined for HW4. At some point they will break off for HW4 data differences and potential performance improvements.
Based on your statement how are Tesla employees testing V11.4 on many cars with HW3 hardware? Many FSD improvements will be coming that have nothing to do with HW3 and HW4 differences.
 
I’d like to see this improvement:

Maybe get into the correct lane sooner (city streets, and sometimes on expressways). The waiting to get into the correct lane 0.1 miles ahead of time is not soon enough. I had the car have to reroute like three times this morning because it couldn’t get into the correct lane soon enough. This has been an issue since 10.2. I’m pretty sure MOST people do not wait until the absolute last few hundred feet to get into the correct lane. It’s mind boggling.
 
But they didn't - dozens of them crashed. AEB didn't even kick in and save them - a standard safety feature of most cars today. Where were the ultrasonics on these cars, or the forward-facing radar for adaptive cruise control?

And where was the common sense from the drivers? A massive dust storm with near zero visibility, and people are thinking "sure, I can drive in this..."

Sounds like it happened all of a sudden. They say ~30 commercial vehicles and 40 to 60 passenger cars. They say dust storms are rare for the area. There was no shortage of professional tractor trailer drivers that got caught up in the melee so it could've happened to anyone.

It would be interesting to hear how many vehicles were equipped with radar/AEB and how well they were able to respond versus the vision/AEB or non AEB equipped but common sense says things can go well with AEB until you get hit from behind from weak links in the chain.
 
  • Like
Reactions: scottf200