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:
Exactly! Annoying how Musk continues to mislead people about FSD. Can’t be by accident, can it?
Its a question I've wondered about ... why does he keep saying that even when proven wrong for many years in a row. Why do people who interview him not ask this question ... ? Probably a topic for the Elon Musk thread. His rational probably goes something like this - "I thought we will get there 4 years back - now we are so much closer, so why not this year" !

Anyway, no one believes him at this point - and in this ER his mumbling on this was so feeble I doubt he believes it all that much either.
 
  • Like
Reactions: Ramphex
This is a valid criticism and I don't really know why they haven't paid attention to that.

Infact we had a whole thread on that ... and I was arguing its a defect and @Terminator857 was saying it is not ;)

Oh, I was referring to me jerking off the yoke to satisfy the FSDj steering nag, while the car is watching me through the cabin camera to make sure I’m watching the road and nothing else.

I find the steering nags extremely annoying and having to look away from my surroundings to check the dash, because just resting my hand on the yoke doesn’t satisfy it, is like constantly playing a game of whack-a-mole, and gets old quick.

If FSDj is so good, then why is there two layers of supervision? Give me smooth ride from that Tesla FSD video, and no forsaken steering torque nags, and I’ll stop bitching about the junk being junk.
 
I may be wrong, but I think Ramph is talking about the wheel nag - having to apply slight torque to the wheel periodically. The "jerk" you alluded to was an actual physics term for g-forces.
Oh, I was referring to me jerking off the yoke to satisfy the FSDj steering nag, while the car is watching me through the cabin camera to make sure I’m watching the road and nothing else.

I find the steering nags extremely annoying and having to look away from my surroundings to check the dash, because just resting my hand on the yoke doesn’t satisfy it, is like constantly playing a game of whack-a-mole, and gets old quick.

If FSDj is so good, then why is there two layers of supervision? Give me smooth ride from that Tesla FSD video, and no forsaken steering torque nags, and I’ll stop bitching about the junk being junk.

Oh - that is a NHTSA thing. I don't think Tesla is getting rid off this anytime soon. It has reduced a lot in 11.x, though ...

BTW, interestingly the nag frequency increases when its raining.
 
  • Like
Reactions: FSDtester#1
I think people need to remember that their FSDj experiences aren’t necessarily the same as the they are for other people in different locations with different models. There are too many inconsistencies that make it hard to compare, almost as we’re comparing different versions of iPhone with different versions of iOS. Yea, the basic concepts are the same, but the behavior isn’t.
An excellent point. I would even add that there has been different behavior in the same location with the same model and the same software version. People have reported their car behaved differently on the same route on a different day or subsequent drive the same day.
 
An excellent point. I would even add that there has been different behavior in the same location with the same model and the same software version. People have reported their car behaved differently on the same route on a different day or subsequent drive the same day.
Yes, this has been stated in this thread many times by many people, over time. A system that is wholly unpredictable is unreliable/not dependable.
 
Oh, I was referring to me jerking off the yoke to satisfy the FSDj steering nag, while the car is watching me through the cabin camera to make sure I’m watching the road and nothing else.

I find the steering nags extremely annoying and having to look away from my surroundings to check the dash, because just resting my hand on the yoke doesn’t satisfy it, is like constantly playing a game of whack-a-mole, and gets old quick.

If FSDj is so good, then why is there two layers of supervision? Give me smooth ride from that Tesla FSD video, and no forsaken steering torque nags, and I’ll stop bitching about the junk being junk.
This has been a constant observation from many on both sides. There are tons of YouTube videos of people who can satisfy the torque requirement with a slight movement from a single finger. And on the other side there are people who describe having to fight the steering wheel just to disengage Auto Steering when an incorrect maneuver is attempted by Beta.

My thought is that the former is the actual intended experience (and one that I have), and the latter is a mechanical issue with the wheel. Especially when I read people who have the car ask for torque, and the do so, and it asks again until it Red Wheels and disengages, even with the driver moving the wheel back and forth. There is something wrong with their steering wheel, or the linkage/motors, which must be causing this. It may be tough to convince service, but a wheel/linkage/steering motor replacement could solve it.
 
Let people know your experience, but you don't have to repeat the same thing over and over again.

Also, just an observation: The people with poor experiences are much more vocal and acerbic on the forums than the so-called shills. You and others have adopted "junk" as your moniker and use it as much as possible, but you rarely see someone with a positive experience with the opposite behavior - such as using "perfection" (FSDp) every post and posting pictures of gold stars or something similar.

Now, if this is all just "fun" and you guys are legitimately bored, then I'd remind you that TMC is growing and new members are joining frequently, some needing help.
Some complaints do get repeated too much.
As far as the balance of positive and negative reports, anything infuriating tends to stick in the mind longer than something pleasing.
If your TV changed channels on it’s own you’d remember that more than how crisp the color is.
Even more so when real danger is involved like lunging or PB.
 
Yes, this has been stated in this thread many times by many people, over time. A system that is wholly unpredictable is unreliable/not dependable.
As long as the system is following basics, and not operating with random intent, then what we may be seeing is the nuances of neural networks. Given 100% exact (only possible in a simulated environment) stimuli, the car may behave the same every time. However, in the real world there are different stimuli even on the same route each time the car encounters it. How the neural networks react and are trained to react to those stimuli can lead to variations on the drive, both good and bad. It's why there is a constant need to train the nets with new samples.
 
As long as the system is following basics, and not operating with random intent, then what we may be seeing is the nuances of neural networks. Given 100% exact (only possible in a simulated environment) stimuli, the car may behave the same every time. However, in the real world there are different stimuli even on the same route each time the car encounters it. How the neural networks react and are trained to react to those stimuli can lead to variations on the drive, both good and bad. It's why there is a constant need to train the nets with new samples.
Perhaps... Some stuff has been bad for nearly two years now, since the first release. With over 400,000 now in the beta program, one would expect to see faster neural net learning than we have been seeing.
 
Last edited:
Perhaps... Some stuff has been bad for nearly two years now, since the first release. With over 400,000 now in the beta program, one would expect to see faster neural net learning and we have been seeing.
If the system was designed to train itself, I'd agree (and also be panicking as we welcome our new overlords), but the system still requires humans to do the work. They have to filter through 400,000 people's experiences, label them, decide which are worthy, and feed them into a training system. I personally think the 400,000 is just Tesla releasing Beta to more people from pressure to deliver, and not about training. The realistic number of samples that the human team can process and train on is likely a fraction of the total.
 
If the system was designed to train itself, I'd agree (and also be panicking as we welcome our new overlords), but the system still requires humans to do the work. They have to filter through 400,000 people's experiences, label them, decide which are worthy, and feed them into a training system. I personally think the 400,000 is just Tesla releasing Beta to more people from pressure to deliver, and not about training. The realistic number of samples that the human team can process and train on is likely a fraction of the total.
🤷‍♂️
 
  • Like
Reactions: Ramphex
Microsoft even used to serve beer in the company commuter busses… No more, I understand…

Re safety - 1) Crossing over the centerline when making right turns, 2) trying to pass stopped and loading school busses with children all around and lights flashing, 3) ignoring school zone loading speed limits with said speed limit sign lights madly flashing, 4) running red lights when you are the lead car, etc., are not shining examples of a “mind blowingly great” product - as often is reported here and by the YT influencers.
Well, after you mind is blown… you are technically “mindless”, right?
 
I’ve posted a couple of videos here of it doing nonsensical things but overall experience on my refresh isn’t smooth, ton of hesitations, slamming brakes while full speed coming up on cars that are stopped at red lights, lane pin balling, getting out of lanes before up coming turns just to try to squeeze back into them or miss the turn and then reroute. Just complete outright nonsense that even an inexperienced driver wouldn’t do. But then there are all these shills with 3/Y posting all these great smooth no intervention for 100,000 mile drives. Starting to get a bit suspect that 3/Y have a better FSDj experience.Maybe

What’s getting old are all the fanboys, influencers and posters claiming how mind blowingly great FSDj is. Now that’s an inaccurate broken record.
Not a fanboy but I am continuously impressed by what it can do. My tag line for my experience is definitely 1% hassle for 99% less stress.

Sorry you don't have the same experience. Maybe you should record some videos and start circulating/ harass tweeting Ashok 😉
 
Or just use the computer software model. You take software with you when you buy a new computer.

That is the model used by SOME software. It's not the model used by ALL software.

If you buy an OEM PC with an OEM version of windows for example, that window license is tied to THAT computer.

Same with OEM Office, and many other examples that are tied to 1 specific piece of hardware- much like FSD is tied to 1 specific car.
 
  • Like
Reactions: FSDtester#1
This has been a constant observation from many on both sides. There are tons of YouTube videos of people who can satisfy the torque requirement with a slight movement from a single finger. And on the other side there are people who describe having to fight the steering wheel just to disengage Auto Steering when an incorrect maneuver is attempted by Beta.

My thought is that the former is the actual intended experience (and one that I have), and the latter is a mechanical issue with the wheel. Especially when I read people who have the car ask for torque, and the do so, and it asks again until it Red Wheels and disengages, even with the driver moving the wheel back and forth. There is something wrong with their steering wheel, or the linkage/motors, which must be causing this. It may be tough to convince service, but a wheel/linkage/steering motor replacement could solve it.
It's my belief based on my experiences with FSDb that Tesla is definitely moving in the direction of eliminating the wheel torque requirement altogether and substituting the interior cameras for monitoring driver attention.

I say this based on how few nags I am now getting with 11.3.6, compared to previous FSDb versions. As long as I'm keeping my eyes on the road ahead and not looking away for a few seconds at a time, I can drive for 3-4 minutes at a time with my hands in my lap, including turns, stops, lane changes, etc. I love this ability!

I think Tesla has taken an interim step with 11.3.6, and I only get the nag every 3-4 minutes if I keep my eyes on the road. This is (to me) a huge improvement over previous releases. I think we'll see a further reduction in torque nags with subsequent releases (assuming our eyes are on the road).
 
This has been a constant observation from many on both sides. There are tons of YouTube videos of people who can satisfy the torque requirement with a slight movement from a single finger. And on the other side there are people who describe having to fight the steering wheel just to disengage Auto Steering when an incorrect maneuver is attempted by Beta.

My thought is that the former is the actual intended experience (and one that I have), and the latter is a mechanical issue with the wheel. Especially when I read people who have the car ask for torque, and the do so, and it asks again until it Red Wheels and disengages, even with the driver moving the wheel back and forth. There is something wrong with their steering wheel, or the linkage/motors, which must be causing this. It may be tough to convince service, but a wheel/linkage/steering motor replacement could solve it.

I don’t know enough about the mechanics to know if the tolerances within the elements involved in the steering and torque detection are big enough to create so vastly different experiences for people. We know that different cars, even within the same model line, have different versions of motors, linkage etc., and that could also affect the experience. Some cars can also have out of spec hardware.

We’ve had few people testing FSD in our car. Most of them complained about how hard it is to find the right amount of force to satisfy the nag. They felt it was either too easy, or too hard, or different each time. My wife, who uses FSD sporadically, struggles with finding an easy way to dismiss the torque nag. It is possible her different driving position, with arms hanging in the air makes it less convenient for her. My arms are on arm rests, and I find it easy to gradually apply more and more torque with either left or right hand to dismiss the nag.

Same car, same roads, different experiences.
 
This has been a constant observation from many on both sides. There are tons of YouTube videos of people who can satisfy the torque requirement with a slight movement from a single finger. And on the other side there are people who describe having to fight the steering wheel just to disengage Auto Steering when an incorrect maneuver is attempted by Beta.

My thought is that the former is the actual intended experience (and one that I have), and the latter is a mechanical issue with the wheel. Especially when I read people who have the car ask for torque, and the do so, and it asks again until it Red Wheels and disengages, even with the driver moving the wheel back and forth. There is something wrong with their steering wheel, or the linkage/motors, which must be causing this. It may be tough to convince service, but a wheel/linkage/steering motor replacement could solve it.
Long time back I used to wiggle the wheel but it would not register. Then I read (somewhere here) that we shouldn't wiggle but gently pull the wheel in one direction hold for a second. That has solved the issue. Whenever I get the nag now, I automatically gently pull the wheel and the nag goes away.

Ofcourse not sure how it works with Yoke .... and with so many cars, obviously there will be mechanical issues in some.