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

 

[aronth5]

And one can't read it unless one has an account. Effectively paywalled.

A free Twitter account is all you need

 

[Tronguy]

A free Twitter account is all you need

Um. Yesterday, Arstechnica published a list of all the things Twitter collects from one’s phone. It’s amazing: may as well hand over everything about myself, medical care, financial details, and what-all to those guys. I’ll check tomorrow and see just how a locked-down Firefox browser can handle it.

 

[Dennisis]

Um. Yesterday, Arstechnica published a list of all the things Twitter collects from one’s phone. It’s amazing: may as well hand over everything about myself, medical care, financial details, and what-all to those guys. I’ll check tomorrow and see just how a locked-down Firefox browser can handle it.

Yeah, quit Twitter over all the crap. Just read, not worth it now.

 

[clydeiii]

Um. Yesterday, Arstechnica published a list of all the things Twitter collects from one’s phone. It’s amazing: may as well hand over everything about myself, medical care, financial details, and what-all to those guys. I’ll check tomorrow and see just how a locked-down Firefox browser can handle it.

How Threads’ privacy policy compares to Twitter’s (and its rivals’) I see nothing medical related here. Also ditch the app and only browse via web, it will limit the data it can see.

 

[PR0FESSOR]

HW4 11.4.4; is anyone seeing ego change routes at the last second and then do a “mad max” maneuver to follow the “new” route? I can get my HW4 11.4.4 vehicle to do it every time on a certain stretch of rural road.

Joe

 

[fasteddie7]

HW4 11.4.4; is anyone seeing ego change routes at the last second and then do a “mad max” maneuver to follow the “new” route? I can get my HW4 11.4.4 vehicle to do it every time on a certain stretch of rural road.

Joe

Yes yes yes!!! there is a road that has a light and going straight and right end up at the same Place, similar ETA. If the light is red or a car there it will suddenly switch lanes with no warning to turn and aggressively accelerate ignoring the no turn on red.

 

[PR0FESSOR]

So… on rural roads where a stop sign is not indicated on the map, FSD 11.4.4 will drive right through the stop sign, ever, single, time. My rural roads are 50 - 55 mph and FSD 11.4.4 will treat the stop sign as if it is not even there. If I force the car to slow down to, say, 20 mph, FSD 11.4.4 sees the stop sign at “the last second” and hits the brakes. A few days ago, it did this twice on the same drive. I always send something to Tesla When FSD does this.

I am almost to the point where I will not use FSD 11.4.4 anyplace but on the interstate. As advertised, it sometimes does the exact wrong thing at the exact wrong time; I feel like I am playing with my life more than I am comfortable with; and according to my DNA on Ancestry, I am in the highest adrenaline junky category.

Joe

 

[PR0FESSOR]

@fasteddie7 In my case, one route is shorter, and is the route I take when driving myself, but the FSD planner does not seem to see the shorter, better, route until the last second… as in, until it is maybe 6-10 meters from the turn. It is as if a light bulb comes on in the deranged mind of FSD 11.4.4 and it says “Oh! There is a shorter route, let’s take it! Then it mashes the brakes and takes the turn so quickly that it throws me sideways.

I don’t let FSD do this if there is a car following me since the sudden braking is truly dangerous on a 55 mph road, but in the back of my mind, I wonder how many other places exist where FSD will do this. I cannot possibly anticipate every possible failure, which actually puts me on edge unless I have seen FSD “behave itself” on a particular stretch of road. Every time FSD updates, I distrust it until I have seen it perform on the road I am driving on.

Joe

 

[rlsd]

FSD 11.4.4 has problem with left turns when there is a car going straight in the opposite direction.

1. FSD was waiting for left turn at the intersection. There were 2 cars at the opposite corner, 1 car waiting for left turn, 1 car waiting to go straight.
Green light was on. FSD moved to the middle of intersection. It correctly waited for the car in the opposite direction to make left turn then continued to make left turn while the car going straight was getting close to its path. I had the impression FSD was going to hit the car going straight. Too much risk. I disengaged. FSD should have stopped in the middle of the intersection to wait for a clear path.

2. While FSD was in the middle making left turn, another car appeared. That car was far away. FSD had plenty of time to finish the left turn but suddenly it turned the wheel back to the right to get back to the original lane. I jerked the wheel and completed the left turn myself. The car then kept moving forward fast forward because TACC was still on after FSD was cancelled when I jerked the wheel. 2 dangerous situations happened at the same time.

 

[FSDtester#1]

Helpful thread even though Telescope says they are just speculating on some of their answers.

https://twitter.com/x/status/1677764913778003968

My S Plaid in for service, they gave me a MSLR as a loaner. It is on 11.3.6.
Beta is enabled and all my cloud profile info came over as well as the app works with the loaner. 11.3.6 in the past 2 days of driving is better than 11.4.4.

 

[FSDtester#1]

FSDj, just being FSDj…

Jeb, will they finally fix my front end vibration from 30 to 48 mph????

 

[LSDTester#1]

Jeb, will they finally fix my front end vibration from 30 to 48 mph????

You betcha

 

[Tronguy]

How Threads’ privacy policy compares to Twitter’s (and its rivals’) I see nothing medical related here. Also ditch the app and only browse via web, it will limit the data it can see.

You're right. Now, Threads apparently can collect medical, private information, at least right now. I just read about that on the first page of the article while my hair turned white(er) and, no, not right now .

I finally got on to pp. 3, and got the following:

Data used to track you​

• Purchases (Purchase History)
• Location (Precise Location, Coarse Location)
• Contact Info (Email Address)
• User Content
• Browsing History
• Identifiers (User ID, Device ID)
• Usage Data (Product Interaction, Advertising Data)

Data linked to you
Third-party advertising:

• Purchases (Purchase History)
• Location (Precise Location, Coarse Location)
• Contact Info (Email Address)
• Browsing History
• Identifiers (User ID, Device ID)
• Usage Data (Product Interaction, Advertising Data)
• Diagnostics (Performance Data)

Developer's advertising or marketing:

• Purchases (Purchase History)
• Location (Precise Location, Coarse Location)
• Contact Info (Email Address)
• User Content
• Browsing History
• Identifiers (User ID, Device ID)
• Usage Data (Product Interaction, Advertising Data)

Analytics:

• Purchases (Purchase History)
• Location (Precise Location, Coarse Location)
• Contacts
• User Content (Photos or Videos, Audio Data, Other User Content)
• Search History
• Browsing History
• Identifiers (User ID, Device ID)
• Usage Data (Product Interaction, Advertising Data)
• Diagnostics (Crash Data, Performance Data)

Product personalization:

• Purchases (Purchase History)
• Location (Precise Location, Coarse Location)
• Contact Info (Email Address, Phone Number)
• Contacts
• User Content
• Search History
• Browsing History
• Identifiers (User ID, Device ID)
• Usage Data (Product Interaction, Advertising Data)
• Diagnostics (Crash Data, Performance Data, Other Diagnostic Data)

Other purposes:

• Location (Precise Location, Coarse Location)
• Contact Info (Email Address, Name, Phone Number)
• Contacts
• User Content (Photos or Videos)
• Search History
• Identifiers (User ID)

Data not linked to you​

Third-party advertising:

• Other Data

Developer's advertising or marketing:

• Other Data

Analytics:

• User Content (Emails or Text Messages)
• Other Data

App functionality:

• Contact Info (Physical Address)
• User Content (Emails or Text Messages)
• Other Data

So, I get it. If one isn't paying for the product, then one is the product. That's how Twitter/Facebook/Threads/et. al. work. The rest (like Mastodon) may be free, but it's volunteer work supported by contributions.

And I still remember Cablevision's terms of service: On one paragraph, they say, "Protecting your privacy is important to us! We make a point of anonymizing your data so nobody can track you."

Then, several paragraphs later, they state, "We run de-anonymizing services." What?

I run with very few Javascripts turned on. And use an ad blocker. So, Twitter? Guess they have to make their money, and making money means taking as many moves as I make on the Internet down on their clipboard and selling it to as many people far and wide as they can. I would object if somebody physically followed me around downtown and watched my every move; why shouldn't I object to Twitter and the like?

 

[FSDtester#1]

You're right. Now, Threads apparently can collect medical, private information, at least right now. I just read about that on the first page of the article while my hair turned white(er) and, no, not right now .

I finally got on to pp. 3, and got the following:

So, I get it. If one isn't paying for the product, then one is the product. That's how Twitter/Facebook/Threads/et. al. work. The rest (like Mastodon) may be free, but it's volunteer work supported by contributions.

And I still remember Cablevision's terms of service: On one paragraph, they say, "Protecting your privacy is important to us! We make a point of anonymizing your data so nobody can track you."

Then, several paragraphs later, they state, "We run de-anonymizing services." What?

I run with very few Javascripts turned on. And use an ad blocker. So, Twitter? Guess they have to make their money, and making money means taking as many moves as I make on the Internet down on their clipboard and selling it to as many people far and wide as they can. I would object if somebody physically followed me around downtown and watched my every move; why shouldn't I object to Twitter and the like?

I have an installed VPN on my laptops and phones, on top of a VPN at the router on top of using this app below. Strongly recommend it. This is the android version, I am sure there is an apple version. It gives you the ability to opt out of data collection and the app was created to combat all these people who want our data. It is not a pay app, it was mandated to help everyone protect their data.
Highly recommend ..Within 6 months of using this, Google search of my public information dropped to page 4 or 5.

AppChoices - Apps on Google Play

Control interest-based advertising on apps on your mobile device.

play.google.com

 

[FSDtester#1]

You betcha

How did they fix yours? Jeb and I have been vibrating for almost 2 years!

 

[Twiglett]

Looks like the release notes for 11.4.5 are identical to 11.4.4, so not likely to see any meaningful improvement of lane discipline

 

[PR0FESSOR]

I teach a class on communications technology from a social science perspective, and I am guessing few of you understand the actual depth of the databases that are being built on you. Every bit of information you give, or share, online, can be used to identify you. The regression algorithms that are being built with the data you provide, can know you better than you know yourself.

Yes, better than you know yourself.

You know what you think you like, who you think you are, what you want others to believe about you. You may think you are putting up a smoke screen, fooling the algorithms. You may fool people, but you will not fool the predictive algorithms. Why? Because what you say about yourself is one thing, but what you actually do with your time online likely says much more about you. For example, you hide your political leanings, but you share your Spotify playlists (or apple music, or whatever). It turns out that your music playlist is around 92% accurate in predicting your political party. All our regression algorithms need to do is find the instrumental variables that represent what they want to know, and it is as good as knowing the real thing. Music preferences tell them your political party and there are literally million of bits of information that we willingly provide which have similar predictive values.

What…. You thought that Facebook asked you your music preferences so that they could offer you music suggestions? LOL. Sure, to the extent that they can sell this information, why not. But Facebook wants it for much more important purposes.

Layer all of the social media platforms, the online forums, Amazon, google, safari, chrome, Microsoft…. And on, and on, and on…

every
single
word
you
type

every
millisecond
you
hover
your
mouse

ever
purchase
you
make

tells those who wish to sell us who we are. They are tracking you, they are selling you, and they can tell you what you will do before you know yourself.

Those of you who read 1984, who think that big brother is a screen that watches you, just like in the book, are wrong.

Big brother is here, now, and he will never go away, because big brother is you and me, trading our information, and our lives, for convenience and the occasional “like”.

I am a public figure and I don’t particularly care what companies know about me, but for private citizens who care about their privacy, sorry, you can run, but you cannot hide.

This said, we drive a Tesla, and we are giving information away at a rate greater than any other population on the face of the planet. What does it matter if Amazon can predict to the exact microsecond, when @Ramphex will have his next hemorrhoid.

Joe

 

[jebinc]

Jeb, will they finally fix my front end vibration from 30 to 48 mph????

Not a F’n chance. Zero probability given these are low volume models.

@WilliamG thinks they will, I heard!

 

[Tronguy]

I teach a class on communications technology from a social science perspective, and I am guessing few of you understand the actual depth of the databases that are being built on you. Every bit of information you give, or share, online, can be used to identify you. The regression algorithms that are being built with the data you provide, can know you better than you know yourself.

Yes, better than you know yourself.

You know what you think you like, who you think you are, what you want others to believe about you. You may think you are putting up a smoke screen, fooling the algorithms. You may fool people, but you will not fool the predictive algorithms. Why? Because what you say about yourself is one thing, but what you actually do with your time online likely says much more about you. For example, you hide your political leanings, but you share your Spotify playlists (or apple music, or whatever). It turns out that your music playlist is around 92% accurate in predicting your political party. All our regression algorithms need to do is find the instrumental variables that represent what they want to know, and it is as good as knowing the real thing. Music preferences tell them your political party and there are literally million of bits of information that we willingly provide which have similar predictive values.

What…. You thought that Facebook asked you your music preferences so that they could offer you music suggestions? LOL. Sure, to the extent that they can sell this information, why not. But Facebook wants it for much more important purposes.

Layer all of the social media platforms, the online forums, Amazon, google, safari, chrome, Microsoft…. And on, and on, and on…

every
single
word
you
type

every
millisecond
you
hover
your
mouse

ever
purchase
you
make

tells those who wish to sell us who we are. They are tracking you, they are selling you, and they can tell you what you will do before you know yourself.

Those of you who read 1984, who think that big brother is a screen that watches you, just like in the book, are wrong.

Big brother is here, now, and he will never go away, because big brother is you and me, trading our information, and our lives, for convenience and the occasional “like”.

I am a public figure and I don’t particularly care what companies know about me, but for private citizens who care about their privacy, sorry, you can run, but you cannot hide.

This said, we drive a Tesla, and we are giving information away at a rate greater than any other population on the face of the planet. What does it matter if Amazon can predict to the exact microsecond, when @Ramphex will have his next hemorrhoid.

Joe

Yep, and I know all that. So:
Ad blockers.
Disable Javascript by default. Enable for just those that I do want.. and that, of course, gives the bad guys an In.

Doesn't help when ISPs monitor internet traffic and collect data. At least T-Mobile gives one a, "don't collect my data" option. Of course, there's the question of, "Do they actually stop collecting data, or lie about it?" Yeah, the FTC could in theory do damage if they lie, but exactly how many times has the FTC landed on a corporate liar?

As you correctly point out, there's no way to completely, or even partially, block all of the dofuses out there selling data. But where it doesn't take much effort to make their lives more difficult, that I do.

As far as Tesla: I actually did read their disclaimer. And, unlike most disclaimers, there's relatively few weasel words, like, "partners". Now, there was a recent fump-de-fump about Tesla workers looking at people's videos: But it was a fump-de-fump because those actions actually went directly against Tesla's notes on the subject. Other places, there would've just been a shrug with an attached, "What do you expect?"

 

[jebinc]

I teach a class on communications technology from a social science perspective, and I am guessing few of you understand the actual depth of the databases that are being built on you. Every bit of information you give, or share, online, can be used to identify you. The regression algorithms that are being built with the data you provide, can know you better than you know yourself.

Yes, better than you know yourself.

You know what you think you like, who you think you are, what you want others to believe about you. You may think you are putting up a smoke screen, fooling the algorithms. You may fool people, but you will not fool the predictive algorithms. Why? Because what you say about yourself is one thing, but what you actually do with your time online likely says much more about you. For example, you hide your political leanings, but you share your Spotify playlists (or apple music, or whatever). It turns out that your music playlist is around 92% accurate in predicting your political party. All our regression algorithms need to do is find the instrumental variables that represent what they want to know, and it is as good as knowing the real thing. Music preferences tell them your political party and there are literally million of bits of information that we willingly provide which have similar predictive values.

What…. You thought that Facebook asked you your music preferences so that they could offer you music suggestions? LOL. Sure, to the extent that they can sell this information, why not. But Facebook wants it for much more important purposes.

Layer all of the social media platforms, the online forums, Amazon, google, safari, chrome, Microsoft…. And on, and on, and on…

every
single
word
you
type

every
millisecond
you
hover
your
mouse

ever
purchase
you
make

tells those who wish to sell us who we are. They are tracking you, they are selling you, and they can tell you what you will do before you know yourself.

Those of you who read 1984, who think that big brother is a screen that watches you, just like in the book, are wrong.

Big brother is here, now, and he will never go away, because big brother is you and me, trading our information, and our lives, for convenience and the occasional “like”.

I am a public figure and I don’t particularly care what companies know about me, but for private citizens who care about their privacy, sorry, you can run, but you cannot hide.

This said, we drive a Tesla, and we are giving information away at a rate greater than any other population on the face of the planet. What does it matter if Amazon can predict to the exact microsecond, when @Ramphex will have his next hemorrhoid.

Joe

Thrombosed, I heard…