FSD Beta v11.x

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.

Deep Understanding Tesla FSD Part 1: HydraNet

From Theory to Reality, Analyze the Evolution of Tesla Full Self-Driving

saneryee-studio.medium.com

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).

Twelve Questions about Tesla Autopilot Technologies Exposed at 2021 AI Day

On Aug. 19 2021, Telsa held an AI day , exposing more technologies about Autopilot , from perception, planning, data labeling, simulation…

yuhuang-63908.medium.com

 

[FOOLSLFDRVNG]

Same. And 11.4.7 is SO good at it, it's predictively slowing down for traffic that won't be on the road until tomorrow.

Yes, our old friends "Phantom Braking" and "Phantom Slowing" have come back for a special guest appearance in the latest release.

They have new sibling twins to join“Phantom Lane Change” and “Phantom Lane Signaling”.

 

[aataskin]

I thought the idea of ditching the radar and the USS was that they don't integrate well with the Vision so each time there are new models of the sensors it would require the whole system to be re-trained:

Of course, the question of why you need to integrate a radar into Vision at all, instead of using it as a fail safe thing to confirm, in real time and more reliably than any Vision ever will, that you are, or are not, about to hit something, was never really tackled (at all, as far as I can tell, never seen anything reasonable written on it).

But now it looks like the Vision itself is not particularly transerable between different models:

Elon Musk Confirms No FSD For HW4 Vehicles (Perhaps Until 2025) as Tesla Prioritizes Solving FSD on HW3 Globally

Elon Musk has confirmed that Tesla will not start work on enabling FSD on HW4 vehicles until the EV maker solves level 5 autonomy on HW3 vehicles globally. Although great for HW3 vehicle owners, this could push FSD access on HW4 until 2025.

www.torquenews.com

Does anybody else feel that the development of FSD is being delayed by Musk's own promises to past customers? I mean, what engineer would refuse to use USS for parking or radar for driving, or having cameras in the nose of the car?

 

[MindBlend]

Any feedback if the new maps are improving FSD?
Version NA-2023.20-14566

Likely depends on where you live. Areas with updated maps do tend to preform much better. But no guarantee your area has updated maps with lane information.

 

[aronth5]

Likely depends on where you live. Areas with updated maps do tend to preform much better. But no guarantee your area has updated maps with lane information.

True but it would be interesting to hear from owners who have the new maps if they see a difference? Maps get pushed out without the owner being notified so it's reasonable to think they only get pushed to areas with map changes. But like everything else with Tesla who really knows.

 

[Dennisis]

I thought the idea of ditching the radar and the USS was that they don't integrate well with the Vision so each time there are new models of the sensors it would require the whole system to be re-trained:

Of course, the question of why you need to integrate a radar into Vision at all, instead of using it as a fail safe thing to confirm, in real time and more reliably than any Vision ever will, that you are, or are not, about to hit something, was never really tackled (at all, as far as I can tell, never seen anything reasonable written on it).

But now it looks like the Vision itself is not particularly transerable between different models:

Elon Musk Confirms No FSD For HW4 Vehicles (Perhaps Until 2025) as Tesla Prioritizes Solving FSD on HW3 Globally

Elon Musk has confirmed that Tesla will not start work on enabling FSD on HW4 vehicles until the EV maker solves level 5 autonomy on HW3 vehicles globally. Although great for HW3 vehicle owners, this could push FSD access on HW4 until 2025.

www.torquenews.com

Does anybody else feel that the development of FSD is being delayed by Musk's own promises to past customers? I mean, what engineer would refuse to use USS for parking or radar for driving, or having cameras in the nose of the car?

Looks like all the HW4 types can forget about FSD if they have to wait for HW3 folks to get to level 5 first. I'm guessing the bar will be lowered at some point .

 

[MindBlend]

Looks like all the HW4 types can forget about FSD if they have to wait for HW3 folks to get to level 5 first. I'm guessing the bar will be lowered at some point .

FSD Beta was released for HW4 last week. Tesla Surprise: Tesla Rolls Out FSD Beta for Hardware 4.0 Vehicles
I'm not sure what any of you are talking about.

 

[Dennisis]

FSD Beta was released for HW4 last week. Tesla Surprise: Tesla Rolls Out FSD Beta for Hardware 4.0 Vehicles
I'm not sure what any of you are talking about.

Referring to post above with Elon's comments. Disconnect somewhere apparently:

• #11,282

 

[AlanSubie4Life]

Referring to post above with Elon's comments. Disconnect somewhere apparently:

• #11,282

All he said is that the software would lag by six months. Fortunately, at the current rate of progress, that means HW3 and HW4 will be indistinguishable in performance!

And subsequently the rollout seems to have confirmed this.

Also anything Elon says can be taken with a lot of skepticism. It’s just random sales babble a lot of the time. You also have to be careful about jumping to conclusions like that article did. Have to very carefully parse a lot of the time, even if it happens to be a truthful statement in a particular instance.

 

[mtndrew1]

Someone was interviewing the charlatan some time back, I think prior to 10.69, and asked him about why FSD hadn’t materialized.

He said something to the effect of “we kept moving towards false horizons” and then said they had to start all over from square one each time.

I’m guessing v10 and v11 have been false horizons and v12 is the next horizon that they’re hoping pans out.

As a related aside I dropped my car off for service last week and had to do a drive with the mechanics to demonstrate my concern. They asked about my new car and I said that I just bought it for the tax credit and FSD transfer; I didn’t really want another Tesla.

I mentioned that FSD has never worked and likely never will work and then they visibly relaxed and laughed with me. Even they know it’s a ruse. I said I only see the potential resale value as EAP and they kind of implied even that was generous.

Anyway, on to the next false horizon.

 

[JulienW]

All he said is that the software would lag by six months. Fortunately, at the current rate of progress, that means HW3 and HW4 will be indistinguishable in performance!....

Also this may have been in context to the development of V12 which is NOT even ready to be tested on HW4 yet. This is one of Elon's biggest problems is assumes everyone knows the complete "backstory" of his one sentence Xest.

 

[Dennisis]

All he said is that the software would lag by six months. Fortunately, at the current rate of progress, that means HW3 and HW4 will be indistinguishable in performance!

And subsequently the rollout seems to have confirmed this.

Also anything Elon says can be taken with a lot of skepticism. It’s just random sales babble a lot of the time. You also have to be careful about jumping to conclusions like that article did. Have to very carefully parse a lot of the time, even if it happens to be a truthful statement in a particular instance.

Yes, a bit tongue in cheek of course.

 

[mtndrew1]

Also this may have been in context to the development of V12 which is NOT even ready to be tested on HW4 yet. This is one of Elon's biggest problems is assumes everyone knows the complete "backstory" of his one sentence Xest.

It’s almost certainly in the context of “please buy one of the eleventy bazillion HW3 inventory cars and don’t worry about finding one with HW4. It’ll def be just as good, pinky swear.”

Tesla will drop HW3 like a hot potato the moment they can legally satisfy whatever was promised on their website at the time of original delivery.

 

[JulienW]

It’s almost certainly in the context of “please buy one of the eleventy bazillion HW3 inventory cars and don’t worry about finding one with HW4. .....

Only the Model 3 comes with HW3, so if you buy any other new Tesla it will have HW4.

As soon as we (US/Fremont) get the refresh (Highland) it will be all HW4 also.

 

[mtndrew1]

Only the Model 3 comes with HW3, so if you buy any other new Tesla it will have HW4.

As soon as we (US/Fremont) get the refresh (Highland) it will be all HW4 also.

Yes, I was referring to the pre refresh HW3 model 3s that Tesla is trying to unload.

 

[sleepydoc]

With adding all the "rules" in code we are just running up against computational power limits.

Yeah, the center lane cannot turn on red, but the rightmost lane can. this sign is one of tens of thousands on the roads that FSDb is going to need to be able to understand if they hope to pull off autonomous driving.

that's the question - aside from the technical aspects of driving we've seen just how large the variations are in road design, signage, etc, not to mention driving styles across the U.S. Creating a system that works in S.F. is hard enough to make one that works in S.F, Dallas, Chicago, Boston and rural Arkansas becomes a colossal challenge. If you go simply on rules you almost need to have different sets of rules for each locale.

 

[sleepydoc]

They have new sibling twins to join“Phantom Lane Change” and “Phantom Lane Signaling”.

and phantom light flashing. On my way to work this morning I actually quit using FSD because the brights were flashing so much (about ever 15-20 seconds they would flash on only to turn off 5 seconds later)

The more we see of 11.4 the more of a dog it's becoming. Just put us back on 11.3 already!

 

[Tronguy]

Someone was interviewing the charlatan some time back, I think prior to 10.69, and asked him about why FSD hadn’t materialized.

He said something to the effect of “we kept moving towards false horizons” and then said they had to start all over from square one each time.

I’m guessing v10 and v11 have been false horizons and v12 is the next horizon that they’re hoping pans out.

As a related aside I dropped my car off for service last week and had to do a drive with the mechanics to demonstrate my concern. They asked about my new car and I said that I just bought it for the tax credit and FSD transfer; I didn’t really want another Tesla.

I mentioned that FSD has never worked and likely never will work and then they visibly relaxed and laughed with me. Even they know it’s a ruse. I said I only see the potential resale value as EAP and they kind of implied even that was generous.

Anyway, on to the next false horizon.

Of course, from what Elon was saying, a better phrase would be, "local maximum".

Those are found in Engineering all the time. When one runs into one, it's time to take a step back, look over the larger problem, and see if there's a better way.
The point of the V12 demo was that, apparently, the crowd at Tesla had found a new and different way of doing things and that the benefits of using that approach resulted in better CPU performance. When Musk was talking about being able to get rid of 500k lines of C++ code; that's code that doesn't need to run. And using NN, according to them, had better performance.

Did you people catch the bit where Musk was repetitively stating, "This drive is so much smoother!" (Of course, that gets in the way of the people claiming the car will have jerky driving from now until forever..)

Is that the end of improvements? Probably not. This whole FSD-b is a research project.

But saying, as people do around here, that, "It's the Great Charlatan Talking!" and "It's a ruse - it'll never work!" makes the naysayers feel so much better.

 

[KirillK]

I've received FSD Beta v11.4.7 few days ago and I can confirm that FSD Beta still doesn't understand the concept of Express Lanes (I-405 in Seattle area).
When I activate it while driving in Express Lane, Tesla tries to exit to regular lane at the first opportunity instead of driving in Express Lane. Toll roads enabled, HOV lanes enabled in settings.

Folks in other states with Express Lanes - are you experiencing the same?

 

[mtndrew1]

Of course, from what Elon was saying, a better phrase would be, "local maximum".

Those are found in Engineering all the time. When one runs into one, it's time to take a step back, look over the larger problem, and see if there's a better way.
The point of the V12 demo was that, apparently, the crowd at Tesla had found a new and different way of doing things and that the benefits of using that approach resulted in better CPU performance. When Musk was talking about being able to get rid of 500k lines of C++ code; that's code that doesn't need to run. And using NN, according to them, had better performance.

Did you people catch the bit where Musk was repetitively stating, "This drive is so much smoother!" (Of course, that gets in the way of the people claiming the car will have jerky driving from now until forever..)

Is that the end of improvements? Probably not. This whole FSD-b is a research project.

But saying, as people do around here, that, "It's the Great Charlatan Talking!" and "It's a ruse - it'll never work!" makes the naysayers feel so much better.

For a very long time I hung on to Elon’s every prognostication about automated driving. This goes all the way back to the 2017.26? branch that was SILKY SMOOTH! (it wasn’t).

FSD clearly is a ruse in that they advertised and sold a product with vivid emphasis and repetitive bolstering for six years and it still hasn’t arrived. Today we’re at HW3 + “at least six months” and I don’t know what else to call that beyond a ruse. They could have very easily tested FSDb in exactly the same manner they do now, for free, and sold the product when it was done.

Much like the hyper loop or the boring company (remember bolting extra wheels onto the side of a car with Jay Leno?), FSD is a technical achievement pretending to solve a problem. That’s neat provided your customers haven’t devoted their resources towards your promise. Maybe gluing extra wheels to the side of a Chevy Bolt and stuffing it into a hole will be common some day, but there’s a reason they’re not selling Boring Bolt Hole Passes right now.

I’m of the mind that, like Twitter, Elon’s enthusiasm got the best of him at the worst time and he simply didn’t understand the first principles of the problem. But then, like Twitter, he was too far in and became mired in his own nonsense.

I also believe that he just doesn’t drive very much, maybe less than a thousand miles a year. And when he does drive it’s on bespoke software that is truly a technical revelation in some respects while being so far from complete that the technical breakthroughs give him a false sense of a timeline. He sees something remarkable, extrapolates, moves on to something else, and ends up at a “false horizon.”

I don’t know what else to call that but a charlatan selling a ruse. I don’t believe he set out in that manner but after the first five or six horizons he clearly knows better. And I fell for it all, hook, line and sinker.

 

[PR0FESSOR]

I suspect that Elon has been practicing “fake it till you make it” for so long, that he doesn’t actually understand how to stop faking it… or perhaps he thinks that he can always fake it. One need look no further than Elizabeth Holmes to see how dangerous it is to fake it a bit too much. This said, America needs Elon Musk, and he will never (I predict) be prosecuted for faking it (including lying and misleading investors the way Elizabeth did). The man is as untouchable as a human being can be, and the only way to get him to change will be to hit him in his pocketbook by not making the same mistake twice (or more).

Joe