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:
So I just drove 4.4 in daytime traffic, and in some ways it's better than 4.2, like moving into the turn lanes in advance, but it has some regressions on a turn across a large bike lane into my neighborhood, so I'd have more disengagements on my commute. 4.2 would more consistently make this turn, so 4.2 would have fewer disengagements although it's not as smooth as 4.4

4.4 also made some stupid decision to overtake some slow cars when our right turn was 40 feet ahead... So it missed the turn obviously

Personally, I prefer 4.4 as a driver assist for its smoothness, but 4.2 would prob have a lower disengagement rate.
 
Just got 11.4.4. First drive on Beta had a small PB, which is unusual on city streets at 45MPH. Also had an odd lane change that said was following route, even though it didn't need to do so. Ended up changing back about .5 miles later.

Also had a URT at a busy intersection. Waited for some time, which was good to see as I've read others mention it just darts ahead and forces a disengagement. It did find a small gap and took it, accelerating aggressively. Was surprised.
 
11.4.4

6-7 miles logged this evening to and from favorite mtb spot. Mind, bike on the back of an upright hitch mount bike rack. So, that’s a factor based on previous experience with a bike on the back. (Not good)

11.4.4 was WORSE than 11.4.3. Absolutely freaked out while making a upl with a truck at said intersection. Also, stopping mid turn just like it did on 11.4.3 on numerous turns. Another “added feature” was absolutely horrible lane keeping on my way home, totally hugging the centerline, never did this before.

I’m taking this with a “grain of salt”, as I had a bike carrier hanging off the back of my car, so, the jury is still out on this. Tomorrow I have some regular traveled routes planned for work, so we will see. First impressions are disappointing (again, I’m blaming it on the bike rack, but who knows).

Who knows, maybe it’s time for a new model s with a roof top bike carrier? Change my mind! But I guess all the refreshed s/x’s suffer from the FSD flu? Yes? No!
 
So torn on this one. Left turns are flawless. Red lights and stop signs are flawless. I was so pumped then the crazy stuff. It went from wanting to cross the line to turn to straight going to ride the shoulder for five car lengths, then made an odd lane choice but corrected itself before the really bad left turn. Then it finally took the left fork but remakes out afterwards. I can see this build on some places being so close to perfect, but there’s some odd stuff in there too.
 
11.4.4

6-7 miles logged this evening to and from favorite mtb spot. Mind, bike on the back of an upright hitch mount bike rack. So, that’s a factor based on previous experience with a bike on the back. (Not good)

11.4.4 was WORSE than 11.4.3. Absolutely freaked out while making a upl with a truck at said intersection. Also, stopping mid turn just like it did on 11.4.3 on numerous turns. Another “added feature” was absolutely horrible lane keeping on my way home, totally hugging the centerline, never did this before.

I’m taking this with a “grain of salt”, as I had a bike carrier hanging off the back of my car, so, the jury is still out on this. Tomorrow I have some regular traveled routes planned for work, so we will see. First impressions are disappointing (again, I’m blaming it on the bike rack, but who knows).

Who knows, maybe it’s time for a new model s with a roof top bike carrier? Change my mind! But I guess all the refreshed s/x’s suffer from the FSD flu? Yes? No!
Yes
 
  • Like
Reactions: FSDtester#1
For anyone who has not yet seen:

Dan is such a clown. He has no idea what he is talking about at all. Does not understand FSDb at all. It’s comical to watch him debate.
 
Last edited:
The prompt broad rollout of 4.4 only a few days past the weekend is very encouraging that development has accelerated. Whatever has changed is exciting.

For me, still no turn signal when merging onto a major road on an entrance ramp. Merge was late but ok.

Managed a difficult cross merge with a garbage truck very well. Only a first drive but seems an advancement.

Had to disengage on a horrible stone one lane bridge (with a blind left curve in the bridge) when ego was about to attempt entering while 2 opposing cars were already entering.

This would be a case for backing up but stopping was sufficient this time for cars to pass. Manually advanced to toward the center of the bridge when a 3rd car entered (blind curve so could not see me) Rather than back up, opposing car chose to slow to a crawl to try to squeeze past me which forced me (not FSD) perilously close to the stone walls. Success but mapping should just avoid this bridge until it is straightened and widened IMO.

Navigating this charming but horrible little bridge works best on horseback. For cars it is like docking at the space station.

Again, the perhaps big deal is the uptick in release deliveries. YMMV
 
The prompt broad rollout of 4.4 only a few days past the weekend is very encouraging that development has accelerated. Whatever has changed is exciting.

For me, still no turn signal when merging onto a major road on an entrance ramp. Merge was late but ok.
I’m just about certain that you’ll get a map update to fix that, the issues I had with merging and early stops signs suddenly went away overnight with no noticeable software update.
 
  • Like
Reactions: FSDtester#1
Chill … turned on minimize lane changes ?
Re: 4.4 also made some stupid decision to overtake some slow cars when our right turn was 40 feet ahead... So it missed the turn obviously

This is an example of something so very very basic that you should not have to find work around to avoid. It should never do this using any option.
 
I’m just about certain that you’ll get a map update to fix that, the issues I had with merging and early stops signs suddenly went away overnight with no noticeable software update.
Thanks for the comment. My first drive was minutes after the 4.4 download. I’ll report back if the merging turn signal appears.

I sometimes wonder if during a deep sleep certain NN sections are swapped to do a little A B testing.
 
For anyone who has not yet seen:

Dan is such a clown. He has no idea what he is talking about at all. Does not understand FSDb at all. It’s comical to watch him debate.
Dan sounded more unstable than O'Dowd. His political biases have no frontal lobe filtering. And non stop chatter from the lady in the back seat. She apparently sat closest to the microphone, was loud and clear, but yet had little useful to say. Not sure why they felt the need to instrument 2 video views of Ross and Dan. That was a waste of needed bandwidth. I would've preferred seeing a closeup of the UI and a wide angle roadway view.

It was hilarious when approaching scenarios with any activity Dan would try to downplay the possibility of success. He'd say this looks challenging while identifying pedestrians and cars and then afterward overstated FSD capabilities. Good for Dan to take the time and try to present the other side but he wasn't anything close to the best FSD spokesperson.
 
I think you guys should mention which setting your FSD uses when describing your experiences... Seems to make quite a difference.

Just about ready to start another 200km worth of driving on the new version. Looking forward to it. 😁
Chill mode on every video test so far. Now that you mention it, maybe I'll do the same type of drives with each of the mode selected to see what kind of difference they're actually is.
 
Thanks for the comment. My first drive was minutes after the 4.4 download. I’ll report back if the merging turn signal appears.

I sometimes wonder if during a deep sleep certain NN sections are swapped to do a little A B testing.
Um. During all the multitudinous over the air updates I've experienced with Teslas, there's been this bit of a trick that I do when I remember: Reset the car, to a greater or lesser degree, after the update. Usually both with the double-scroll-wheel reset and, if possible, the power-down on the Safety submenu.

There have been cases in the past where Weird Behavior was observed on a newly updated car that got fixed with that.

And, if one squints, there might be data in the car that doesn't get wiped on an update. So, resetting after an update might prevent bizareness. In general, I mean.

Comments?
 
  • Like
Reactions: FSDtester#1
People always say reboot the car after an update.

Which is always weird to me because the car reboots itself after the update.
Sometimes systems have a first_boot flag that causes them to do additional things like initialize databases or other persistent storage. This could cause a delay in one, or more, processors, or processes) becoming fully operational. If another process is expecting a specific boot order, there could be a race condition that causes anomalous behavior. The race condition might work out OK 99% of the time, so may not get detected until the software gets to the field. I've had Tesla updates that acted weird on the initial boot, but cleared up after rebooting.

Given Tesla's rapid development pace, I would not be surprised if bugs like this slip by. So, a second boot might not be a bad strategy.