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:
For anyone stuck on 10.69, it does appear that rebooting the car "fixes" this as per an earlier post (I think on this thread but I'm unable to find it). My Model 3 finally has 2022.45.15, I can stop borrowing my wife's car to try V11 on the freeway.

So far, the freeway performance has been mostly positive. It does seem to switch to the right lane significantly later than 10.69 did. A few times I was little worried that it wasn't going to make it.

My experience with cut-ins seems opposite of what I seen reported here. It seems to "not flinch" in several situations that I was ready to take over. It seems to be willing to use a closer gap for a period of time and then grow it (similar to what I would do in tight situations).

On surface streets I definitely like the moving over to the right to turn right change, it is much closer to how I would drive. Very small sample size so far so I haven't seen any of the weird routing that I've read about.
 
  • Like
Reactions: Dennisis
I have one case where it's definitely related to the width of the lane (two lane road, well-marked and -paved), and another for which I haven't figured out a cause (divided four lane road, well-marked and -paved). As mentioned, Chuck Cook found that it will not enable if you're at the edge of a unmarked road. In my case, FSDb will happily drive on the problem roads if enabled beforehand, but doesn't want to enable on those roads.

Ultimately, it isn't accidental. Tesla is intentionally not allowing FSDb to be enabled if they don't like the current driving environment. Like most things they do, it's too conservative.
There's a spot on my daily commute where I can't enable now on v11 (both.3 & .6). What's interesting and counter to your statement is that if it's enabled before I hit that section, no problem and stays on.

Road is not narrow and clearly marked with yellow center and white edge.

Never had any issue on 10.x.

So Tesla does allow FSD to be enabled there, just not allowing to be initiated.
 
  • Informative
Reactions: pilotSteve
There's a spot on my daily commute where I can't enable now on v11 (both.3 & .6). What's interesting and counter to your statement is that if it's enabled before I hit that section, no problem and stays on.

Road is not narrow and clearly marked with yellow center and white edge.

Never had any issue on 10.x.

So Tesla does allow FSD to be enabled there, just not allowing to be initiated.
Here's a more up to date post about some testing I did on this. It looks to me like it has to do with clutter close to the car's lane.
 
  • Like
Reactions: RowdyMY
It seems to be willing to use a closer gap for a period of time and then grow it (similar to what I would do in tight situations)
Yeah, I've definitely noticed this behavior for freeway onramp merges too. Typically this is when you're merging from the right, but I was pleasantly surprised by 11.3.6's behavior when we were the vehicle in the right-most interstate lane with a plain immediate onramp merge. There was a lead vehicle and a merging vehicle, and FSD Beta actually sped up to close the gap down to 0.5s follow distance behind the lead vehicle so that we would make space for the merging vehicle before growing the gap back to the usual 1.5s for Chill after passing the merge. The merging vehicle and we were pretty close to side-by-side before this quick decision to close the gap.

I suppose it might not be that surprising as there's merges from the left lane ending and the usual merges from the right-side onramp, and both sides should result in similar behavior whether you're in the lane that ends or lane that continues. If there's no distinction to FSD Beta, that could help identify some potential issues if there are mispredictions of upcoming merges and/or these merges shouldn't actually be treated the same?
 
I am currently on hw4 2023.6.11.2 I see 2023.6.15 fsd is rolling out. I currently don't have the FSD request button (greyed out) says available in future update. So I don't make a mistake, if I update to this version, I should then have a request button, then after approval I'll be in, correct?
 

Attachments

  • Screenshot_20230417-114048.png
    Screenshot_20230417-114048.png
    467 KB · Views: 59
I am currently on hw4 2023.6.11.2 I see 2023.6.15 fsd is rolling out
That's very good news for you that it's starting to roll out as FSD Beta 11.4 / 2023.6.15 to employees. Indeed HW4 vehicles are getting updated to 2023.6.11.2 while HW3 vehicles 2023.6.8 and 2023.6.11. All of those version numbers seem like they would be compatible with upgrading to 2023.6.15 although it's not as clear if HW4 will be included in the initial 11.4.x rollout. The most recent rollout of FSD Beta 11.3.x had separate versions for vehicles with and without ultrasonic sensors, but they were combined pretty quickly after individual testing.

A lot of people so far have been unable to get any form of FSD Beta 11.x yet because it's built on an older software version 2022.45.x while many vehicles especially recent deliveries have been on 2023.x software.
 
  • Helpful
Reactions: pilotSteve
V11.4 release notes:


Edit - Text extract:

FSD Beta v11.4
Autopilot Improvements
- Improved the decision to assert or yield for pedestrians at more crosswalks by evaluating multiple possible futures in the joint space of ego's actions and the pedestrian's response.
- Improved ego's behavior near VRUS by measuring their probability of
intersecting ego's path, based on their kinematic data, and
preemptively decelerating when the estimated risk is high.
-Improved turn performance in dense unstructured city environments. Examples of improved cases include: turning when the turn lane is blocked by parked cars and avoiding turning into bus lanes.
-Improved lane guidance module to feed in long range routing "hints" to the network for which lanes ego needs to be in to reach its destination. Also significantly improved per-lane routing type autolabeler. These changes combined resolved 64% of all interventions caused by bad routing type.
-Improved geometric consistency between lane, line, road edge and restricted space detections by re-training our networks on the same dataset with the latest version of our "lane guidance" module, and by using a common features space to predict line, road edge and restricted space.
- Improved recall for partial cut-ins by 39% and precision for false positive cut-ins due to lane changes into adjacent lanes by 66%. resulting in a 33% reduction in overall lane-changing prediction error. This was accomplished by further increasing our auto-labeled fleet dataset by 80k clips, improving the accuracy of the auto-labeling algorithm, and tuning the distribution of training supervision.
- Improved understanding for when to use bus lanes and when to avoid them, by updating the lane type detection network and improving map- vision fusion.
- Improved speed control during lane changes through better consideration of upcoming navigation deadlines, required back-to- back lane changes and presence of a vehicle behind ego.
- Added new Vision Speed network to infer the typical driving speed on a given road. This is used to limit the maximum allowed speed in environments such as parking lots and residential roads.
-Mitigated hydroplaning risk by making maximum allowable speed in Autopilot proportional to the severity of the detected road conditions. In extreme cases, Autopilot may use the wetness of the road, tire spray from other vehicles, rain intensity, tire wear estimation or other risk factors that indicate the vehicle is near the handling limit of the surface to warn the driver and reduce speed.
-Improved long-range path blockage detection and control on city streets. Ego will now be able to perform lane changes due to upcoming path blockages earlier.
-Improved developer productivity with better code diagnostics and C+ +20 features by upgrading compiler to clang-16. This also improved photon-to-control vehicle response latency by 2%.
 
Unfortunately I already have Joe mode on.
You may have disagreed, but as I said, it didn’t know what it was doing, and using it in significantly degraded conditions is dangerous:

“Mitigated hydroplaning risk by making maximum allowable speed in Autopilot proportional to the severity of the detected road conditions. In extreme cases, Autopilot may use the wetness of the road, tire spray from other vehicles, rain intensity, tire wear estimation or other risk factors that indicate the vehicle is near the handling limit of the surface to warn the driver and reduce speed.”

We’ll see if this speed adjustment is concomitant with increased following distance and other adjustments needed for inclement weather. It seems unlikely that this safety improvement is complete (thus it likely remains unsafe except in highly curated situations).

Remember that Tesla provides lots of latitude, and assumes the driver will only use the system when it is safe, and will take care to avoid any instances where FSD will have a collision. This is a tricky balance, but Tesla will typically err on providing more latitude, even if it means an accident caused by the driver.
 
Last edited:
- Improved recall for partial cut-ins by 39% and precision for false positive cut-ins due to lane changes into adjacent lanes by 66%. resulting in a 33% reduction in overall lane-changing prediction error.
Ok, this is one I experienced, when a vehicle turned into an adjacent lane ego went all "scaredy cat" and slowed down. So maybe that will be resolved, or at least better, in 11.4.

I wonder if we are finally to the point where the speed of delivered improvements will accelerate.
 
Ok, this is one I experienced, when a vehicle turned into an adjacent lane ego went all "scaredy cat" and slowed down. So maybe that will be resolved, or at least better, in 11.4.

I wonder if we are finally to the point where the speed of delivered improvements will accelerate.
Yeah I think so. I think the major change took a lot of time behind the scenes to update and now we will get quick updates to improve things based on interventions, disengagements, and voice notes. Exciting times ahead!
 
Tesla seems to be pushing out a not-FSD-Beta software 2023.6.11 that merged those those two versions: with and without USS. Those updating are coming from both those versions as well as those who were on recent 2023.2.100/.200 factory software versions for new 3/Ys.

This potentially is in preparation for a 2023.6.15 / FSD Beta 11.3.6 merging 11.3.4/11.3.5 to the latest production branch allowing FSD Beta additions to recent deliveries and FSD Capability acquisitions already on 2023.x software, which is probably at least 90% of the fleet not already on FSD Beta.

Your crystal ball on this turned out to be amazingly accurate. How did you know that 2023.6.15 was going to be the build that merged FSD with the 2023 branch? Just wondering if there is a published roadmap somewhere out there that I am missing. Thanks for shedding light on it either way...