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:
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.
Again: This is a real, no-kidding Beta we're playing with, here. Nobody should be calling it a final product because, well, it's not. A couple of days ago my copy actually ran a stop sign. (It had shown signs of doing this so, on this go-around, I made darn sure there weren't any approaching cars from other directions, then let it do its thing. Ran right past the stop sign, which was on the display, went 3/4 of the way into the intersection, half-way halted, then jerked its way out of there. Clear fail. Too bad we don't get a "report" button any more.)

We get the thing at the reduced price because it's not ready yet, just like an unbuilt house. Come the day that it is ready, then the car will be worth a heck of a lot more, robotaxies R Us. At which point Tesla will start selling FSD/Robotaxi like it's worth the money people will be making on it when the car's not being driven by the owner, but it running paying passengers, a la Uber, around the landscape.

Tesla's said that, come the day, they're going to crank up the value of FSD to account for the true earning potential of the car (that's their words, not mine). And, when they were talking about all this a couple of years ago, the thought was that Tesla would start running fleets of robotaxies around, making money on the same.

Fine, that's Tesla's line, but, no kidding, they've been sticking to it.

There's plenty of people around here and elsewhere who think that That Day Will Never Come. But there are others who think it may. If it doesn't, it sure won't be due to a lack of trying.

We'll see.
 
When I bought my model S in February 2015, the salesman (associate) mentioned that Musk was talking about a driverless trip from NY to LA happening soon, maybe the same year. It's getting closer for sure. I watch Chuck Cook's test runs and a few of the others. It still makes rookie mistakes that will lead to a collision without driver intervention. It's whack-a-mole with each release. One day, they will whack that last mole, but predicting when that will happen is quite difficult. I'm certain that the stock will pop when that OMG moment happens. I'd be surprised if it isn't a 1.5x to 2.0x bump.
 
When I bought my model S in February 2015, the salesman (associate) mentioned that Musk was talking about a driverless trip from NY to LA happening soon, maybe the same year. It's getting closer for sure. I watch Chuck Cook's test runs and a few of the others. It still makes rookie mistakes that will lead to a collision without driver intervention. It's whack-a-mole with each release. One day, they will whack that last mole, but predicting when that will happen is quite difficult. I'm certain that the stock will pop when that OMG moment happens. I'd be surprised if it isn't a 1.5x to 2.0x bump.
Given the rate of progress and plateaus the past two years, it seems like my that time my Plaid will be over the hill and long gone.
 
  • Like
Reactions: tmoz
11.4.4 is so incredibly bad.
Sorry I just had to type that again.
But seriously, it’s bad.
I can’t go 1 mile without disengaging. We are years away from actual FSD. Actually, I don’t think that with current road design it will ever happen. Think about a gate that requires you to type in a code to open it? How will the car know to stop where you can open the gate?
But before they solve the infinite level of edge cases they have to solve the extremely basic stuff first. I mean, the car can’t even make basic lane selection decisions correctly 80% of the time? How in the world are people saying that this thing is good?
 
  • Like
Reactions: jebinc
11.4.4 is so incredibly bad.
Sorry I just had to type that again.
But seriously, it’s bad.
It handles 95+% of my driving now. Best version so far by far. And most of the time my passenger doesn't know if FSD is driving or not. I'm curious why it seems to be so different for other people.

One thing I found is if I press the accelerator to make it go faster around a corner it goes very poorly. Either let it go the speed it wants or take over and drive yourself.

I mean, the car can’t even make basic lane selection decisions correctly 80% of the time?
Sure, sometimes it makes stupid choices like changing to the left lane only to change back to the right lane 50 yards later. But it does it safely and gets to the destination. (I do sometimes cancel lane changes if they move away from a required exit/turn that is coming up quick.)
 
It handles 95+% of my driving now. Best version so far by far. And most of the time my passenger doesn't know if FSD is driving or not. I'm curious why it seems to be so different for other people.

One thing I found is if I press the accelerator to make it go faster around a corner it goes very poorly. Either let it go the speed it wants or take over and drive yourself.


Sure, sometimes it makes stupid choices like changing to the left lane only to change back to the right lane 50 yards later. But it does it safely and gets to the destination. (I do sometimes cancel lane changes if they move away from a required exit/turn that is coming up quick.)
You guys do watch Chuck Cook's videos, right? Narrow road without striping, the car would either stop or turn towards the on-coming car. It's better now - doesn't turn towards the oncoming car, but often just stops. Unprotected left turns with a safe landing median appear to be solid except for picking the final lane to wind up in. Having said that, I don't doubt for a second that this version is the best version so far in general, but how far away from all bugs being worked out is still a mystery - 3 months, 3 yrs ? It's coming, but it is anybody's guess in my opinion.
 
11.4.4 is so incredibly bad.
Sorry I just had to type that again.
But seriously, it’s bad.
I can’t go 1 mile without disengaging. We are years away from actual FSD. Actually, I don’t think that with current road design it will ever happen. Think about a gate that requires you to type in a code to open it? How will the car know to stop where you can open the gate?
But before they solve the infinite level of edge cases they have to solve the extremely basic stuff first. I mean, the car can’t even make basic lane selection decisions correctly 80% of the time? How in the world are people saying that this thing is good?
Many are not, but rather are calling it what it is; over priced junk.
 
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.

I think you are right. And I think most people would be shocked to learn what low percentage of all that real world driving data gets processed at all. I would be surprised if it's more than a single digit percentage.

The comma.ai guy said that anything more than the data from a few thousands cars is overkill for the autonomous driving work. He could be full of it, but I still bet Tesla looks at very little of it.
 
So what is happening with 11.4.6? 🤔 Did the employee trusted with testing it go and lose it? 🤪 Still no sign of it being released. Seems Tesla is struggling with 11.4.x even more than past versions. How are we going to have L4 or L5 "maybe later this year" if we can't even get a stable version of 11.4.x this summer?:oops:🤣

Now back to waiting and waiting and .......👴
 
  • Like
  • Funny
Reactions: tmoz and jebinc
Seems Tesla is struggling with 11.4.x even more than past versions.
Given the recent statements of a change in approach, I would not be surprised if we experience a period of minor or slipping feature advances. Focus could be shifting from delivering FSD releases toward massive harvesting of driving video of any kind. Nightly pulls from vehicles could go way up.

They are clearly suggesting massive compute is the secret sauce and that is growing but depends on Nvidia in the short term. I am moderating my expectations until ‘24.

2024 could be a real rocket ride as compute ramps.

What I expect is that chaos in the personnel space distracts however. Multimillion dollar signing bonuses driving departures from the FSD team.
 
So what is happening with 11.4.6? 🤔 Did the employee trusted with testing it go and lose it? 🤪 Still no sign of it being released. Seems Tesla is struggling with 11.4.x even more than past versions. How are we going to have L4 or L5 "maybe later this year" if we can't even get a stable version of 11.4.x this summer?:oops:🤣

Now back to waiting and waiting and .......👴
11.4.7 released this weekend - Elon

Joking aside, fixing software bugs is not a quick task. 3-6 week testing is a short time. I think this time Tesla wants to make sure the release really fixes the serious issues. It's a good sign. Tesla cannot afford to fix one bug and introduce a new bug anymore. I hope we will have a decent version by Labor Day. Or it may take longer. They may have to rework the lane keeping, lane changing for route, and left turn, right turn approach.
 
Last edited:
Last night, I was driving back home from work on a rural highway, around 90km/h. It's a 4-lane road and I was driving on the right lane with FSD on, and my hand is on the wheel as well...

It was at dusk (auto lights on), so things weren't very visible. I was obviously paying extra attention. Then all of a sudden, the car started to swerve towards the left lane. I couldn't figure out why, and then, WOW :oops: a person with a buggy (not sure what was in it) was walking on the line towards me! I only saw her when we were very close. The car avoided hitting her because we were already partially off from the center. Crazy that a person would even think about walking at that spot at that time of the day!

I'm not going to claim FSD saved an accident, but clearly, it saw more than I did, at least under the above said condition and acted before what I would have done.
 
Last edited:
Last night, I was driving back home from work on a rural highway, around 90km/h. It's a 4-lane road and I was driving on the right lane with FSD on, and my hand is on the wheel as well...

It was at dusk (auto lights on), so things weren't very visible. I was obviously paying extra attention. Then all of a sudden, the car started to swerve towards the left lane. I couldn't figure out why, and then, WOW :oops: a person with a buggy (not sure what was in it) was walking on the line towards me! I only saw her when we were very close. The car avoided hitting her because we were already partially off from the center. Crazy that a person would even think about walking at that spot at that time of the day!

I'm not going to claim FSD saved an accident, but clearly, it saw more than I did, at least under the above said condition and acted before what I would have done.
Sometimes it does great things, and other times very unsafe things... The point being it can't be trusted - and you ALWAYS have to be on "High Alert" with this product (especially on non highway roads). I'm looking for less anxiety, not more, and hence the reason I'm waiting for Tesla to remove FSDj from my car...
 
Sometimes it does great things, and other times very unsafe things... The point being it can't be trusted - and you ALWAYS have to be on "High Alert" with this product (especially on non highway roads). I'm looking for less anxiety, not more, and hence the reason I'm waiting for Tesla to remove FSDj from my car...
I believe we all know this already - FSD works better on some situations than others, and depending on where you live, quality of your experience will vary by a wide margin. Obviously, you are in a more complex area. For me, I would describe FSD drive (in my driving situations) as focusing 70% of my usual attention when FSD is on, and my hand on wheel is acting as a third eye as a precaution. It helps a lot.

It is ironic since even my wife agrees that this is the best $15k we have ever spent on anything in our lives (not just cars), even in its far less than perfect form. It results in less anxiety, and it is far far less stressful for our driving needs.

But then again, we are getting towards retirement age and we have moved away from a big city (Toronto). Had we not move from that place, our opinion might have been totally different.
 
  • Like
Reactions: Thp3 and tmoz
Sometimes it does great things, and other times very unsafe things... The point being it can't be trusted - and you ALWAYS have to be on "High Alert" with this product (especially on non highway roads). I'm looking for less anxiety, not more, and hence the reason I'm waiting for Tesla to remove FSDj from my car...
We will miss your productive input but will certainly let you know when to return….
 
  • Like
  • Funny
Reactions: tmoz and jebinc
I believe we all know this already - FSD works better on some situations than others, and depending on where you live, quality of your experience will vary by a wide margin. Obviously, you are in a more complex area. For me, I would describe FSD drive (in my driving situations) as focusing 70% of my usual attention when FSD is on, and my hand on wheel is acting as a third eye as a precaution. It helps a lot.

It is ironic since even my wife agrees that this is the best $15k we have ever spent on anything in our lives (not just cars), even in its far less than perfect form. It results in less anxiety, and it is far far less stressful for our driving needs.

But then again, we are getting towards retirement age and we have moved away from a big city (Toronto). Had we not move from that place, our opinion might have been totally different.
Wife and I feel the same. With managed expectations we find it low stress and impressed by the evolution and progress. No it’s not done or perfect but that was not our expectation. Taking a swing at the FSD transfer in the next week as that was the only thing holding me back to progress to the next tier Model S from our MY. (For us) we wouldn’t buy this technology car without the technology. The transfer made that an easy justification.