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:
I wished there was more room for nuance on the internet. FSD is both not what was promised several times in the past, and it’s currently useful to me.

I find it very easy to turn my brain on and off on the task of driving, so even if I’m having to intervene every 5 minutes, I still find utility in that. Maybe it’s my ADHD. I don’t know.
First of all I think it would go without saying at this stage in the evolution of social media that my subjective statements are, of course, my opinion from my perspective. New rule: unless someone is citing an authoritative source, assume any subjective statements are their opinions.

Second, I am quite aware that different people have different experiences. But what makes no sense to me is that those experiences can vary so wildly on the same vehicle with the same software. I’ve have driven on NoA for quite a few years and find it very useful in both open road driving and bumper-to-bumper traffic (not so much in heavy traffic at highway speeds, though, because of continued phantom braking problems).

FSDb (Autosteer on City Streets) is a completely different animal. I have never had an intervention-free drive. The thing can’t make a mile or two without requiring and intervention. And while I can contemplate some reasons why other folks may let it go a little further than I do (e.g., they’re not in a hurry so they let it miss a turn or they don’t care if they are pissing off other drivers around them), it’s hard to imagine how my drives in suburban Atlanta can be so different from other parts of the U.S. that folks are taking it on intervention free drives and/or finding driving with it to be relaxing or useful.

Now I am an engineer so I tend to be very precise and objective, and I am rarely impressed by technology. And I’m sure that affects my viewpoint. But again none of that explains the wild differences in standards that exist when evaluating whether the current version FSDb is “quite good” vs. just “marginally better” than the last version.
 
it’s hard to imagine how my drives in suburban Atlanta can be so different from other parts of the U.S. that folks are taking it on intervention free drives and/or finding driving with it to be relaxing or useful.
I think the solution to this puzzle is: this is not actually happening.

I haven’t seen a single such video of any useful length. Even Whole Mars’ “intervention free” drives have tons of interventions (meaning any normal person would intervene, many many times). There does not have to be an intervention for it to count as an intervention!

Video evidence might change my mind but I certainly have not seen it.

The latest version is definitely improved over prior versions as I said and on occasion can be surprisingly capable, but any normal individual would intervene a lot.

I think everyone’s experiences are basically the same. Certainly sounds like it to me. FSD has come a long way, but has a long way to go.
 
Last edited:
So we went through all of February waiting for Investor Day to be some kind of milestone day for FSDb. ....
Why anybody thought this is beyond me. Very similar to how many thought Tesla would reveal the next gen car on investor day. Not a chance for either.
This is what Tesla sent out and it's reasonably close to what actually happened. "Our investors will be able to see our most advanced production line as well as discuss long term expansion plans, generation 3 platform, capital allocation and other subjects with our leadership team."

If we've learned anything in following Tesla daily since 2009 is that Elon rarely uses major "events" to go into details. And every event is followed by people complaining when he doesn't. Great example is the recent Semi reveal with Frito Lay. Most wanted and assumed more details and very little was provided. Does that drive me slightly crazy yup, but I don't expect it now.

Whether you believe Elon should, he doesn't so get used to it.
 
  • Like
Reactions: powertoold
I think the solution to this puzzle is: this is not actually happening.

I don't have video evidence, but I have had many intervention-free drives on non-trivial routes in the Houston metro area (say typically 20 -ish miles drives with 7 miles of it NoA, several complex intersections, from my suburban driveway to a restaurant parking lot). They're not the norm, but they can happen.

More common is that I have 2-3 mild and reasonably-predictable intervention/disengage events in such a drive, and 2-3 other events that others might intervene for, which I let play out if it seems circumstantially safe to do so (I used to instruct noob track drivers from the passenger seat, so FSD doesn't really scare me much!).

There are also scenarios and areas where I know it will be frustrating and don't even bother trying anymore. It's a mixed bag for sure!
 
I think the solution to this puzzle is: this is not actually happening.

I haven’t seen a single such video of any useful length. Even Whole Mars’ “intervention free” drives have tons of interventions (meaning any normal person would intervene, many many times). There does not have to be an intervention for it to count as an intervention!

Video evidence might change my mind but I certainly have not seen it.

The latest version is definitely improved over prior versions as I said and on occasion can be surprisingly capable, but any normal individual would intervene a lot.

I think everyone’s experiences are basically the same. Certainly sounds like it to me. FSD has come a long way, but has a long way to go.
Yep. It's not a matter of cars behaving differently or some drivers knowing how to operate FSD better. No doubt it's a tough problem. It would be nice if Tesla and Elon just admitted they are still trying to find the pieces to the puzzle.
 
  • Disagree
Reactions: STUtoday
Interesting - is this a different warning than they previously had? It calls out Full Self Driving capability features but not basic Autopilot. Very curious to see if these cars have no autopilot functionality at all. People aren't going to like that. Due to the FSD Beta recall they weren't going to get into Beta immediately anyway.

I wonder what sort of software integration work they have to do. Emulate the previous generation camera suite in software by downscaling (and maybe cropping if the field of view is wider) the new camera feeds. And then run the existing software (but on the HW4 computer).

If that works is delaying any Autopilot functionality on those cars I can totally see why that would be by far highest priority. I'm just a little confused why that wasn't ready by now though - they must have known for many months that HW4 cars would be delivered to customers around now. Did they really run into difficult issues getting the existing AP software working?

It seems overly bold but within the realm of possibility for Tesla that they bet everything on getting FSD V11 working on HW4 (for both basic AP on highways and for city streets) with no intent of getting legacy production AP working. And now if V11 has issues... it could be holding up the whole thing...

That would be classic Tesla...
 
  • Like
Reactions: powertoold
It would be very interesting to see an experiment where more then one Tesla drove the same route at approximately the same time, but not with one car using the other as a lead car. While the conditions would never be exactly the same for both, they would be dealing with the same intersections, lighting, visibility, etc. The question is how repeatable is FSDb from car to car.

If FSDb is not very repeatable between cars, then there may be some physical issue, like camera vibration, suspension wear, etc., affecting performance. FSD is predicated on each car acting the same way. But, do they?

All of the YT videos are lone wolfs. How about it tubers? Care to do a joint video between two or more of you?
 
But that screenshot says "no fsd", so it is definitely something new.
It really doesn't say "no FSD", and I'm not simply being a contrarian. I just think you have to read it more carefully.

It says that "...some features may be "temporarily limited or inactive at delivery including [a few that are part of FSD capability] and Full Self Driving capability features..."

This language is extremely similar, though not quite the same, as the language in the original press release regarding USS removal, last autumn well before any HW4 news. So why do they add the somewhat redundant mention of FSD capability, to a list that already names some FSD capability features? I think partly just due to imprecision, but also possibly to emphasize the point.

To me, it's not even so clear that this "Proceed with Delivery" agreement language is any different from that of non-USS cars delivered since November or so. I found only one other screenshot example, and that had quite similar language but was shorter, with a "Learn More" button suggesting another page that wasn't shown. These web pages are displayed only to people who are about to take delivery.

There certainly has been a strong rumor that HW4 computer-equipped cars needed some software work to become functional and deliverable - but that most likely is a general software backbone problem, as the HW4 module includes a great deal of core functionality beyond autopilot functions, including the now on-board multimedia control functionality.

Whether or not Jason's Tesla is an HW4 Model X, I think it's highly premature to conclude that non-beta FSD is completely missing. This agreement page is about the USS issue and there's no indication it's HW4 specific.
 
  • Helpful
Reactions: Matias
Wonder if the screen "blanking out" on 11.3 is strictly a MCU issue and unrelated to the FSD Computer. So in other words it is not a safety issue and only a matter of getting the UI to display what the car is doing. Here is evidence that it is only UI. The car is coming up an onramp and as it gets near the top it blanks out and the blue tentacle has the car still turning right even though it is now going straight. A couple of frames and the UI comes back on and the tentacle is facing straight.

Screenshot 2023-03-04 at 7.55.13 AM.png
Screenshot 2023-03-04 at 7.55.35 AM.png
 
Last edited:
  • Like
Reactions: KrazinAz
Don't be too envious of HW4 vehicles, yet. It sounds like they're shipping without any FSD features for the moment. Maybe the Autopilot team has delayed work on V11 to work on getting HW4 up and running:

Wow no USS? Seems like a big mistake. Wouldn’t have bought my model x without it. Mine was a 23 demo I picked up last week. And with those added features I paid 9k less 😂. Tesla needs to do a price decrease shipping such an expensive car without it
 
Wonder if the screen "blanking out" on 11.3 is strictly a MCU issue and unrelated to the FSD Computer. So in other words it is not a safety issue and only a matter of getting the UI to display what the car is doing. Here is evidence that it is only UI. The car is coming up an onramp and as it gets near the top it blanks out and the blue tentacle has the car still turning right even though it is now going straight. A couple of frames and the UI comes back on and the tentacle is facing straight.
Yes. Either an MCU issue or an FSD computer issue with the code the prepares and/or sends visualization data to the MCU. You can clearly see the car continuing to drive with no warnings. Still, an issue that needs to be addressed.
 
I don't have video evidence
If you don’t have video Alan won’t believe it.

I have regular intervention free-drives, too. (Albeit without video so these are also mythical) I also have plenty of drives where the only intervention is to goose the accelerator because I don’t want to inconvenience the cars behind me when I’m making a turn, or to switch lanes because I’m happier in a lane that has fewer potholes.
I understand the caution and not rolling something out too quickly. They are under a microscope right now, so rushing anything out could backfire. I'd rather they take their time and release something a bit more cautiously.

I'm even preparing myself mentally for the possibility that the car will start reading more speed limit signs and adjusting to them - such as school zones. On a Sunday evening my car may suddenly slow down to 25MPH from 45MPH when it sees a school zone sign. Better to be safe than try to guess if there are kids somewhere around the school. Then we, as the driver, can override and wheel up the speed if we deem it legal to do so.
You’re probably right about being under a microscope - Tesla obviously knew about the NTSB investigation and impending recall far earlier than the public did. The other issue is that the current version of Auto Pilot on highways is very good, near perfect in my experience. That gives them a pretty high bar for the new version. People using FSDb generally understand and expect it to make mistakes but if Tesla releases a significant regression of the highway capabilities there will be a lot of anger amongst the masses.

As far as speed limit signs, the school zone issues is virtually impossible for humans to deal wtih - I’d be happy with a warning beep saying “reduce speed for school zone?” As it is, the current version actually seems to be worse than past versions at recognizing regular speed limit signs.

v11 appears to be a major rewrite and it’s almost impossible to do that without introducing new bugs so rationally I’m expecting a regression yet I still find myself hoping for improvements.
 
Here is evidence that it is only UI. The car is coming up an onramp and as it gets near the top it blanks out and the blue tentacle has the car still turning right even though it is now going straight. A couple of frames and the UI comes back on and the tentacle is facing straight.
Which illustrates to me how frightening I would find FSD without the visualization.

I seem to recall the engineers at AI day saying that the FSD stuff has no operating system per se. Instead, all the scheduling is determined in advance (compile/build time). If they have a fixed time per pass and the calculation goes long, that suggests that they'd have to just end the calculation at some point. If so, then the last steps should be the least critical, and I'd think that communicating with the HDU would qualify. That FSD code may be slow/unoptimized for a number of reasons related to it still being in development, occasionally making it run long, compromising those low priority tasks at the end. This is just speculation, of course.