Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

FSD V11 Discussion - First Released 11/11/22 at 11:11PM - Maybe

This site may earn commission on affiliate links.
(moderator note: Added maybe to the title. This may be an alpha or beta release, so “release” is relative. Stay tuned.)

We finally have confirmation of the release of FSD V11. This presumably is single stack; we’ll see, as the question was left unanswered. I guess it will still be called FSD Beta V11?

002905FC-FC4B-4346-9970-AF2BA0758CE4.jpeg


🤯 ( 🔥 )
 
Last edited by a moderator:
I hope they do, and I could see a lot of reasons to do so. They have a robust incremental rollout strategy for FSD updates (developers, all Tesla employees, small external beta group / long time beta testers, newer beta testers, all users). This is exactly how you want to roll out software that is complex to test and is safety critical.

It also gives them user groups to do AB testing on. I'm not sure Tesla has ever confirmed this for sure, but there's a lot of anecdotal experiences where different testers on the same version see a surprisingly large variation in performance (consistent with some runtime tuning that is different).

After the full public wide release of FSD for city streets and single stack highway AP, they've still got a lot of development to go. Offhand:

- Parking lots
- Driveways and garages (really anything that's not a public road)
- Any kind of maneuvers that require backing up
- Inclement weather
- Slippery surfaces (snow and ice)
- Locations beyond US + Canada
- and just incremental improvements on everything FSD can do today to make it more reliable and work towards L3/L4 operation
- (maybe) some kind of HW4 upgrade with cross traffic cameras further forward on the vehicle (presumably would require some retraining and testing)
I understand what you say and would normally be in agreement. However, Tesla does not use public beta groups for the production software. I believe that Tesla took the step to invite FSDb testers because they needed the video and telemetry data from a large number of scenarios. Now that FSDb has been used over a year, they may feel that they have sufficient data for testing future releases and additional data can be gathered from the FSD-enabled fleet at large. Note that many users have reported that their cars upload very little data now. And, the current 10.69.3.1 release has no video snapshot button. This indicates, to me, that Tesla can now set specific triggers to silently capture and send needed data with no user involvement beyond doing their normal driving.
 
I'm not sure Elon realizes that Thanksgiving is tomorrow? I'm hoping we get to see some YouTubers try it out soon, but "before Thanksgiving" would have to be today:

You’re not missing much, 36.20 is one of those two steps back releases for me at least. Car has done things it never did before, in a bad way. Strange sudden jerks to the right of the road (like into the bike lane) as if it was trying to avoid a car or dog or something - and yet nothing is there, not even shadows.

Pulling out onto my no traffic street we‘re back to acceleration that is close to breaking traction. Unnecessary, obvious not situational at all or varied. I get it, the car has to be able to accelerate for the crazy Florida unprotected left over a median situation, but my feeling is that same sort of execution does NOT need to be applied EVERYWHERE.

Why put on blinkers while driving down a straight street? I don’t know either.
 
I think you and I have gone back and forth on this a couple times already.

so no further reply is warranted.
Ok. Ideally we’d measure the g-forces, just to get an idea. Breaks traction around 0.8-0.9g with the worst tires. Good tires are more like 1.1-1.2g as I recall, roughly. All assuming a decent & clean road surface of course.

get it, the car has to be able to accelerate for the crazy Florida unprotected left over a median situation,
Unfortunately it can do this but it does not do it in any software to date. About 20-30% slower than a typical human right now.
 
  • Like
Reactions: Yelobird
Ok. Ideally we’d measure the g-forces, just to get an idea. Breaks traction around 0.8-0.9g with the worst tires. Good tires are more like 1.1-1.2g as I recall, roughly. All assuming a decent & clean road surface of course.


Unfortunately it can do this but it does not do it in any software to date. About 20-30% slower than a typical human right now.
But without exaggeration it doesn’t sound as impressive. Lol
 
On a total aside, I was reading an article on Aviation and Automation. Apparently, regulators are pushing for more automation to help reduce the load on pilots (and potentially eliminate co-pilots for some flights).

I recall many discussions about the term "autopilot" in Tesla vs the aviation industry. And this observation from a pilot was an eye opener, and drew parallels to Tesla and some of the issues AP/FSD have had:
Wait .. is this the autopilot Dan O'Dowd works on ?
 
I don't think that anyone will need to request city streets FSD once it is released wide at the end of the year. It will still be 'beta' like all of Tesla ADAS features, but it will no longer be 'early access beta'. No safety score will be required. All you will need is have an FSD license (purchase or subscription) and to enable it like one does currently for AP or NOA.

With the removal of the camera icon and the Autopilot Buddy defeat, I really do wonder whether you might end up being right here.

I definitely need to prepare to have my mind boggled. It may just happen.

I still think it won’t happen by the end of the year (I guess that would make me right), but Elon is really throwing caution to the wind these days, so who knows.

I figured wide release would be mid-2023 at the earliest based on current behaviors.

I still disagree with people who think NOA will be changed. That doesn’t make sense since it is a different product. Will just run FSD stack but have all the same options (maybe there will be a release where that is not the case, due to time constraints, I guess we will see). But end state seems like it should be the same as currently since there isn’t any reason to change the product that EAP owners have.
 
I understand what you say and would normally be in agreement. However, Tesla does not use public beta groups for the production software. I believe that Tesla took the step to invite FSDb testers because they needed the video and telemetry data from a large number of scenarios. Now that FSDb has been used over a year, they may feel that they have sufficient data for testing future releases and additional data can be gathered from the FSD-enabled fleet at large. Note that many users have reported that their cars upload very little data now. And, the current 10.69.3.1 release has no video snapshot button. This indicates, to me, that Tesla can now set specific triggers to silently capture and send needed data with no user involvement beyond doing their normal driving.

I disagree that they won't have a need for testing with a larger public beta fleet. "they may feel that they have sufficient data for testing future releases" - that doesn't quite make sense to me. The beta program isn't just about collecting video snippets they use for training; a core piece of Tesla's strategy is rapidly iterate and get new functionality into the hands of real world testers. Old data is of limited use for, say, testing if a change to the planner leads to fewer interventions, or any other change to driving behavior.

"However, Tesla does not use public beta groups for the production software." - that's true, historically, but they also haven't shipped safety critical changes directly to the production group (changes like Vision Only for highway AP were piloted with the FSD program and then ported over). I just can't see why Tesla would use a robust incremental release strategy for 2+ years of development and then just stop! They'd just be taking on unnecessary risk by abandoning incremental rollouts. That decision would be all downside and no upside! FSD Beta isn't anywhere near being done in the sense that they don't have substantial code changes to make on it in the next year. Releasing a cut of it to the production fleet is just one step in the broader FSD program.

Anecdotally, my car has uploaded over 60GB of data since it got 10.69.3.1 (no video snapshot button, that's all automatic telemetry) so data collection is well and alive.
 
The beta program isn't just about collecting video snippets they use for training; a core piece of Tesla's strategy is rapidly iterate and get new functionality into the hands of real world testers.
What do call the hundreds of thousands of FSD owners? Once FSDb goes wide distribution, Tesla can get data from every one of those cars that has data sharing enabled. They couldn't do that before because the software was seen as too immature to release to the pubic at large.
but, once the software goes to everyone, it can report back every time a driver takes control. Tesla can even specify triggers to cars to cause it to silently record snapshot data if the car encounters certain scenarios.
 
What do call the hundreds of thousands of FSD owners? Once FSDb goes wide distribution, Tesla can get data from every one of those cars that has data sharing enabled. They couldn't do that before because the software was seen as too immature to release to the pubic at large.
but, once the software goes to everyone, it can report back every time a driver takes control. Tesla can even specify triggers to cars to cause it to silently record snapshot data if the car encounters certain scenarios.
I understand all of that. I feel like you're missing my point about the risk mitigation benefits of incrementally releasing driving behavior changes.
 
I understand all of that. I feel like you're missing my point about the risk mitigation benefits of incrementally releasing driving behavior changes.
I'm not missing your point. I believe that Tesla has enough current vehicle data, and has developed the capability to task vehicles for more as needed to support simulation capability. They may also be able to deploy test versions of FSD to run in the background in parallel with the production software and perform tests without users even being aware of it. That mitigates risk better than deploying software to untrained testers.
 
  • Like
Reactions: AlanSubie4Life
So, (some) employees don’t even have v11 yet. My cousin is an employee and always has the latest and greatest (she had 10.13 back when that came out). She’s on 10.69.3.1 just like the rest of us, and she got that update the same day I did. So much for the inside track?
 
I have a theory and would love @EVNow Or @Bladerskb opinions.
Or any engineer or programmers opinion.
When it goes to single stack, can or would they potentially Delete/Uninstall
any NOA or Highway code thereby freeing up much needed storage and compute resources?
If FSDb is just running the show on all roads, they could dump any other programming no longer needed, right or wrong?