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:
Yea, I just don't see a ton of evidence that FSDb has been tested much at all on the highway.
Good point!

Some thoughts on this issue: The standard AutoPilot-on-highways has been around for a long time and has a well established safety record. It does have eccentricities which I, for one, long to see improved upon. But it seems to me that before rolling out an entirely new version based on the FSDb code, Tesla would want to be pretty darn sure that the new version is at least as safe and capable as the old one.

"Safety" is generally measured in small numbers of crashes and deaths per million miles of driving. Statistics 1A says it will take many, many millions of miles of driving on the new version to be certain that is is, in fact as safe as the old. This would seem to argue for a phased roll-out: simulation, in house testers, employees, slowly enlarging volunteer customers, etc. Each phase gives better statistical certainty while minimizing the risk of harming the testers or other drivers. Because the important testing has to be done on public roads, the public shares the risk. So a very cautious roll-out seems necessary.

I also wonder about the effects of higher speeds in the highway environment will affect the performance of the FSD code. What works at lower speeds may simply not work, or need substantial adjustments. For example, higher speed means it must see problems at greater distances, and reactions must be gentler, making a larger volume in which objects must be recognized and tracked, and the responses must be different than when at lower speeds. The way FSD yanks sometimes on the steering wheel would put a car into an uncontrollable twizzle-spin if not a roll-over, not a good maneuver on a crowded interstate. One of those caught on a dash cam might well end the entire FSD project.

So, just maybe, there are multiple practical issues which make "single stack" V11 extra tricky to get to market. Tesla had it a bit easier when the fleet was small and growing. But a premature release to even the current FSDb testers could be a disaster.

Like I said, these are just some thoughts. I do expect it will work, but for now I am trying to rally my patience.

SW
 
Good point!

Some thoughts on this issue: The standard AutoPilot-on-highways has been around for a long time and has a well established safety record. It does have eccentricities which I, for one, long to see improved upon. But it seems to me that before rolling out an entirely new version based on the FSDb code, Tesla would want to be pretty darn sure that the new version is at least as safe and capable as the old one.

"Safety" is generally measured in small numbers of crashes and deaths per million miles of driving. Statistics 1A says it will take many, many millions of miles of driving on the new version to be certain that is is, in fact as safe as the old. This would seem to argue for a phased roll-out: simulation, in house testers, employees, slowly enlarging volunteer customers, etc. Each phase gives better statistical certainty while minimizing the risk of harming the testers or other drivers. Because the important testing has to be done on public roads, the public shares the risk. So a very cautious roll-out seems necessary.

I also wonder about the effects of higher speeds in the highway environment will affect the performance of the FSD code. What works at lower speeds may simply not work, or need substantial adjustments. For example, higher speed means it must see problems at greater distances, and reactions must be gentler, making a larger volume in which objects must be recognized and tracked, and the responses must be different than when at lower speeds. The way FSD yanks sometimes on the steering wheel would put a car into an uncontrollable twizzle-spin if not a roll-over, not a good maneuver on a crowded interstate. One of those caught on a dash cam might well end the entire FSD project.

So, just maybe, there are multiple practical issues which make "single stack" V11 extra tricky to get to market. Tesla had it a bit easier when the fleet was small and growing. But a premature release to even the current FSDb testers could be a disaster.

Like I said, these are just some thoughts. I do expect it will work, but for now I am trying to rally my patience.

SW
Whatever problems exist with V11 must be pretty significant as there has been no word of updates since the original 11 Nov releases. No public testers seem to be involved yet. If V11 was about to be released, we would expect to have seen multiple releases to beta testers so that Tesla could get widespread test data from interstates across the country in a diverse set of traffic and weather conditions. We've not seen any of that yet.

Musk said there would be a release of V11 this week. We'll see, but I suspect that the 2022.44.25.5 was supposed to be that release. But V11 wasn't ready, so they went with 10.69.25 as plan B.
 
  • Like
Reactions: AlanSubie4Life
Musk said there would be a release of V11 this week. We'll see, but I suspect that the 2022.44.25.5 was supposed to be that release.

The way things have been going, I expect Elon to overrule everyone and just tell them to YOLO it in the next couple days. He’ll accompany it with a Tweet about how it would be a crime not to release it. And they’ll provide a toggle to turn it off (until such a toggle disappears, there is not really a true single-stack release - it’s just some experimental software that is less safe).
 
Last edited:
  • Like
Reactions: pilotSteve
Did Elon say it would be a release to customers? Perhaps an internal release? Or, perhaps it was just Elon as usual leaving out the "not before" qualifier as usual. Or perhaps another Emissions Testing release, aka vaporware? ;-)
Internal doesn’t count as a release. That is called internal testing. Release means to the customer base.
 
Curious that the updates for FSD Beta 10.69.3.3 / 2022.40.4.10 and FSD Beta 10.69.25 / 2022.44.25.5 have been so different from previous releases. The former went to new FSD Beta testers increasing the TeslaFi FSD Beta fleet by about 5% while also going to a slightly smaller number of existing testers on 10.69.3.1, and it went first to "not-11" employees then Early Access group and new testers at the same time. The latter so far is roughly a 5% rollout, and more interestingly that it went to "not-11" employees then existing Safety Score group skipping those in Early Access:

Could this different rollout behavior be in preparation for something FSD Beta 11 related?
 
Curious that the updates for FSD Beta 10.69.3.3 / 2022.40.4.10 and FSD Beta 10.69.25 / 2022.44.25.5 have been so different from previous releases. The former went to new FSD Beta testers increasing the TeslaFi FSD Beta fleet by about 5% while also going to a slightly smaller number of existing testers on 10.69.3.1, and it went first to "not-11" employees then Early Access group and new testers at the same time. The latter so far is roughly a 5% rollout, and more interestingly that it went to "not-11" employees then existing Safety Score group skipping those in Early Access:

Could this different rollout behavior be in preparation for something FSD Beta 11 related?
One part of me wants to believe that there's some V11 thing about to drop, but I still think that V11 won't be seen until next week at the earliest. So, I expect the floodgates will soon open on 10.69.25 and we'll all be pushed that very soon. It is strange that the Early Access group didn't get it first, but, if this is really 10.69.3.3 dressed up in a Santa suit, then there's really nothing new to test FSD-wise. I haven't seen any 10.69.25 videos on youtube, so it certainly appears that the usual crowd has not gotten first dibs.
 
  • Like
Reactions: edseloh
One part of me wants to believe that there's some V11 thing about to drop, but I still think that V11 won't be seen until next week at the earliest. So, I expect the floodgates will soon open on 10.69.25 and we'll all be pushed that very soon. It is strange that the Early Access group didn't get it first, but, if this is really 10.69.3.3 dressed up in a Santa suit, then there's really nothing new to test FSD-wise. I haven't seen any 10.69.25 videos on youtube, so it certainly appears that the usual crowd has not gotten first dibs.
Yea, I'm just not seeing any V11 based build going out to any cars