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

 

[FSDsjw]

Not to offend anyone, but Mars is more a shill than an objective beta tester. FSD is always amazing in his eyes.

Probably ok for some to be relentlessly optimistic posters. A lot of you who have been on FSD beta for years probably don’t remember the wonder of seeing the FSD d its thing for the first time. I just got the Beta and am really in awe right now. Understand of course that most of the frustration is related to promises of a future state that have been very slow to come.

 

[Twiglett]

Hard to be too optimistic after Elon's false claim that V11 was released on 11/11 at 11 even though it wasn't. For me until we see it actually released am going to remain skeptical. V11/single stack has had more false perceived starts, imminent release timelines and pullbacks in the last year than almost any Tesla/Elon proclamation (excluding Full Self Driving that is coming before the end of A year).

indeed - I'm feeling much more cynical about so called FSD recently because of the constant broken promises and price increases, but the product itself really is getting better (mostly) at each release.

 

[JulienW]

indeed - I'm feeling much more cynical about so called FSD recently because of the constant broken promises and price increases, but the product itself really is getting better (mostly) at each release.

I am more and more also. While Beta is improving if I were to scale it from last Oct v10.2 to today and what is needed for L4......well.... Let's say we started at 1 on 10.2 (only YouTubers had it before so that is where I have to start) and L4 is 100. I would rate 69.25 about 20 to 30. So while lots of improvements to go from 1 to 20 it is still YEARS away from getting to 100. And the closer it gets to 100 the (exponentially) harder it will be to make progress. It is hard to see a "magic bullet" happening in 2023 that moves us soooo much farther along. Hope I'm wrong or mis-valuing the progress we have made so far.

 

[EVNow]

I am more and more also. While Beta is improving if I were to scale it from last Oct v10.2 to today and what is needed for L4......well.... Let's say we started at 1 on 10.2 (only YouTubers had it before so that is where I have to start) and L4 is 100. I would rate 69.25 about 20 to 30. So while lots of improvements to go from 1 to 20 it is still YEARS away from getting to 100. And the closer it gets to 100 the (exponentially) harder it will be to make progress. It is hard to see a "magic bullet" happening in 2023 that moves us soooo much farther along. Hope I'm wrong or mis-valuing the progress we have made so far.

As I keep writing, human level is 1 in 10,000 miles (accidents). Waymo is also at a similar level in terms of disengagement. But crowdsourced disengagement number is about 1 in 10 miles. NOA is about 1 in 100 miles. So, NOA needs to get 1 to 2 orders of magnitude better and FSD needs to get 2 to 3 orders better.

 

[dramsey]

As I keep writing, human level is 1 in 10,000 miles (accidents). Waymo is also at a similar level in terms of disengagement. But crowdsourced disengagement number is about 1 in 10 miles. NOA is about 1 in 100 miles. So, NOA needs to get 1 to 2 orders of magnitude better and FSD needs to get 2 to 3 orders better.

But you're comparing accidents to disengagements; the two aren't the same. A disengagement doesn't equate to an avoided accident.

 

[JulienW]

But you're comparing accidents to disengagements; the two aren't the same. A disengagement doesn't equate to an avoided accident.

Just 30 minutes ago on my way home Beta tried to blow through a Stop sign and there was traffic coming. So it was a disengagement BUT had I not it MAY have been an accident.

So in my mind Beta is even farther behind that accident statistic would indicate.

 

[boonedocks]

Some of these youtubers make a lot of money from youtube videos .... enough to buy them a Y or even an X. So they produce videos that gets them max views ...

I sure don’t sponsor any of them.

I am pretty certain some of us here put many MANY more FSDBeta miles on our cars than the YouTubeShillers do and don’t get any recognition or special attention from Tesla. (Early access).

I find a great number of them don’t even show the real world antics of FSDB. It they did they would probably be tossed to the end of the pre-release software line

 

[momo3605]

As I keep writing, human level is 1 in 10,000 miles (accidents). Waymo is also at a similar level in terms of disengagement. But crowdsourced disengagement number is about 1 in 10 miles. NOA is about 1 in 100 miles. So, NOA needs to get 1 to 2 orders of magnitude better and FSD needs to get 2 to 3 orders better.

Where did you get the statistic that accidents are 1 in 10,000? That sounds wrong to me. From a quick google search, I’m seeing varying stats but seems to be closer to 1 in 100,000 miles.

 

[EVNow]

Where did you get the statistic that accidents are 1 in 10,000? That sounds wrong to me. From a quick google search, I’m seeing varying stats but seems to be closer to 1 in 100,000 miles.

Virginia Tech, modified for urban use by Cruise. We are talking about simple accidents like curbing the wheel, that never get reported to police (or even insurance).

Miles between incidents for regular cars (according to Cruise’s modified version of Virginia Tech study, which removes crashes on highways and rural roads):​

​

Miles between Level 1 incidents: 282,485​

Miles between Level 1 and 2 incidents: 99,404​

Miles between Level 1, 2 and 3 incidents: 20,661​

Miles between Level 1, 2, 3 and 4 incidents: 9,218​

Automated Vehicle Crash Rate Comparison Using Naturalistic Data – Virginia Tech Transportation Institute

Self-driving cars are quickly moving from prototype to everyday reality. During this transition, the question that is first and foremost on the mind of the public and policy makers is whether or not self-driving cars are more prone to crashes. This would seem to be an easy question to answer...

featured.vtti.vt.edu

https://www.reddit.com/r/SelfDrivingCarsLie/comments/bxx7ea

ps : Also ...

 

[dramsey]

Just 30 minutes ago on my way home Beta tried to blow through a Stop sign and there was traffic coming. So it was a disengagement BUT had I not it MAY have been an accident.

So in my mind Beta is even farther behind that accident statistic would indicate.

I've had FSD beta since early June, and it's never blown through a stop sign or red light for me. In fact I can't think of a disengagement I've had that could have been an accident if I hadn't disengaged-- it's generally stuff like "No, that's not the right lane to be in now".

So in my mind, Beta's entirely different.

Sadly there's no way to collect accurate statistics on this, so I suppose the evidence will have to remain anecdotal.

 

[AlanSubie4Life]

so I suppose the evidence will have to remain anecdotal.

There’s also video evidence.

 

[powertoold]

Since Karpathy left, Elon and the FSD team have had larger communication discrepancies.

We used to take Elon's tweeted timelines with a grain of salt; nowadays, his timelines hold no salt at all, lol.

Worse yet, we don't get any technical explanations for the delay. Are they have trouble with training? Or the architecture? Or HW3 limitations? I think the silence on the matter is telling that these technical hurdles won't be solved in a way that leads to a step-change in fsd performance on HW3.

We can see from AI Day 2 that the NN leadership is lacking without Karpathy. Maybe I'm gonna be wrong. Hope I'm wrong.

 

[EVNow]

I've had FSD beta since early June, and it's never blown through a stop sign or red light for me. In fact I can't think of a disengagement I've had that could have been an accident if I hadn't disengaged-- it's generally stuff like "No, that's not the right lane to be in now".

So in my mind, Beta's entirely different.

Sadly there's no way to collect accurate statistics on this, so I suppose the evidence will have to remain anecdotal.

Never happened to me either … but there are enough people who say it has happened to them ….

 

[powertoold]

My 69.3.1 sometimes blows through unprotected rights with mini islands to the left.

As for blowing stop signs, it's never outright blown through them, but it has stopped and then proceeded confidently even though I don't have confidence (because I couldn't see for myself).

 

[AlanSubie4Life]

… but there are enough people who say it has happened to them ….

And posted corroborating videos!

 

[JulienW]

I've had FSD beta since early June, and it's never blown through a stop sign or red light for me. In fact I can't think of a disengagement I've had that could have been an accident if I hadn't disengaged-- it's generally stuff like "No, that's not the right lane to be in now".

So in my mind, Beta's entirely different.

Sadly there's no way to collect accurate statistics on this, so I suppose the evidence will have to remain anecdotal.

Here you go and a couple of bonus including trying to run a red light.

 

[willow_hiller]

As of 5 hours ago, that pending 2022.44.25.20 version was installed, and is now being called V11.X on TeslaFi:

https://twitter.com/x/status/1607661067877990402

Hopefully we get confirmation soon whether it's V11.2 or V11.3, and some updated release notes.

 

[Maxters]

As of 5 hours ago, that pending 2022.44.25.20 version was installed, and is now being called V11.X on TeslaFi:

https://twitter.com/x/status/1607661067877990402

Hopefully we get confirmation soon whether it's V11.2 or V11.3, and some updated release notes.

at this point I just want to see what it looks like.. I've never been so excited for ASS in my whole life.. Also would like the visualizations to change with EAP on the highway.

 

[Mardak]

As of 5 hours ago, that pending 2022.44.25.20 version was installed

Yeah, that vehicle seems to be following its usual behavior when this incoming version was spotted Christmas Eve:

It's been 4 days and still no release notes, and I would guess most "employees" who had the update available installed it on Christmas Day, so Tesla seems to have been successful in keeping things from disclosed for an even larger group of testers. So far just this one vehicle that has typically gotten these before-early-access releases, while another seemingly "employee" vehicle receiving updates early has been removed from TeslaFi (now just shows 0 Odometer and VIN XXX), and the third vehicle getting 10.69.3.3 early does not have this update pending. So potentially even for "employees," this FSD Beta 11.x version isn't available to that whole group yet or people are better at keeping it secret.

TeslaFi.com Firmware Tracker 2022.44.25.20 b65011

 

[Supcom]

Yeah, that vehicle seems to be following its usual behavior when this incoming version was spotted Christmas Eve:

It's been 4 days and still no release notes, and I would guess most "employees" who had the update available installed it on Christmas Day, so Tesla seems to have been successful in keeping things from disclosed for an even larger group of testers. So far just this one vehicle that has typically gotten these before-early-access releases, while another seemingly "employee" vehicle receiving updates early has been removed from TeslaFi (now just shows 0 Odometer and VIN XXX), and the third vehicle getting 10.69.3.3 early does not have this update pending. So potentially even for "employees," this FSD Beta 11.x version isn't available to that whole group yet or people are better at keeping it secret.

TeslaFi.com Firmware Tracker 2022.44.25.20 b65011

The encouraging thing about this vehicle getting V11.x is that it does not appear to have received the earlier V11, indicating that the new version is being rolled out a little wider. Hopefully, we will see a version released to public testers in the near furture.