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

 

[jebinc]

There's still 11.4.4 coming this weekend.

Perhaps… perhaps, not…

 

[fasteddie7]

I believe, with the current release, it just stops.

Ok. Then no, it's not reacting to the no turn on red sign for me.

 

[arnolddeleon]

Given TACC (in theory) slows down with objects ahead to avoid hitting them- how would it be easy to get into an accident if TACC remains on?

I've actually thought about this BTW, because I agree initially it seems weird... but if TACC turns off, you immediately get regen braking without obvious reason, which seems MORE likely to cause an accident by a tailgater than continuing to drive with active cruise on would?

I've also thought about this a little and have to the conclusion that is likely "safe enough". I will tend to flip the lever up to take full manual control. If I don't and it surges I will quickly tap the brakes to take full manual control. If I'm in testing mode and anticipating something a little dicey I will have my fingers stick out to flip the lever to disengage fully as well.

 

[Pdubs]

Same here, but 3 Powerwalls. It could be the same vendor!

Sunrun

 

[jebinc]

Sunrun

No sun here in Seattle!

 

[Yelobird]

No sun here in Seattle!

Seattle in general is one of the most beautiful places abound with rivers, volcanos, wildlife, travel there a Lot. All beautiful for the 5 days a year you can see without rain and clouds. Lol

 

[jebinc]

Seattle in general is one of the most beautiful places abound with rivers, volcanos, wildlife, travel there a Lot. All beautiful for the 5 days a year you can see without rain and clouds. Lol

Everything is a trade off. But, we typically get zero rain from 7/1-10/15. We have only two seasons…. It rains and it doesn’t rain!

 

[jebinc]

Everything is a trade off. But, we typically get zero rain from 7/1-10/15. We have only two seasons…. It rains and it doesn’t rain!

Today…

 

[sleepydoc]

There seems to be some scant evidence it may also read "No right on red" signs in the most recent beta, but this is just anecdotal atm (DirtyTesla has shown it seeming to honor this on 11.4.3, but we shall see...).

With the last few releases I've found FSDb to be inconsistent in taking right turns on red. It could be that these anecdotal reports are actually because of bugs in the right turn on red algorithms. Or it could be that they have a very buggy algorithm to recognize 'no right turn on red' and it's getting it wrong.

 

[sleepydoc]

My first drive on 11.4.3 was really good. I did a 30 mile drive yesterday across Orlando. I engaged in the driveway and disengaged when it stopped at the destination. Didn't touch a thing. So far so good.

Except for the rocking back and forth I get above 50mph... everything was pretty good.
I have to figure out what is causing that. I'll probably have to make a SC appointment.

Is the rocking occurring independent of FSDb or is it from something like FSD wandering within the lane?

 

[jabloomf1230]

With the last few releases I've found FSDb to be inconsistent in taking right turns on red. It could be that these anecdotal reports are actually because of bugs in the right turn on red algorithms. Or it could be that they have a very buggy algorithm to recognize 'no right turn on red' and it's getting it wrong.

My guess is that certain intersections with NRToR have been entered into Tesla's navigation database, based on user reports. Signage is too variable for Tesla to be wasting time on recognition, when there are bigger problems needing immediate solution. Eventually Tesla will get to this.

 

[uscbucsfan]

11.4.3 doesn't seem to be either better or worse than previous versions. It still drives in the center of unmarked roads, only to swerve and stop for an oncoming vehicle.

The car surprised me by waiting patiently and then deciding to pass a stopped garbage truck, gracefully maneuvering over the double yellow line. Then it avoided a barricaded detour which it had previously ignored, despite a wall of orange cones showing on the UI which represented the barricade. But my joy was short lived. The car then approached a green light in order to make a ULT. Instead of stopping, it decided to cut off an oncoming pickup truck. Fortunately, I had just enough time to disengage and turn to the right.

There's still 11.4.4 coming this weekend.

.3 and .4 are bug fixes.

 

[Pdubs]

No sun here in Seattle!

But you have great coffee, music and an amazing fish market.

So there’s that. And everybody I know there eats honey.

 

[swedge]

I am finding the blinkers a bit confusing. Three different circumstances seem to need different inputs.

First, driving manually with blinkers not set to automatically turn off after a lane change. So I start the blinker, full press so it stays on, make my lane change and need to turn it off. It seems I need to press it in the same direction to stop it. True?

Second, now with it set to turn the blinker off after a lane change, sometimes it does not do turn itself off, perhaps after entering the freeway. But sometimes is does. Confusion and distraction occur when I turn it off manually, with a tap in the same direction, but I do that just as it turns itself off. So, my tap actually turns it back on, oops. Any suggestion on how to avoid this?

Finally, and here is where FSD comes in, the car is driving and decides it wants to change lanes. It gives me a chance to cancel. Should I tap the blinker, full tap the blinker, or tap in the other direction?

I know I could go out driving and experiment to figure out its logic, but freeways here are too fast and crowded to drive along randomly blinking, blinking. ;-) Actually, around here, FSD testers are the only one who do signal for lane changes, though I doubt other driver mind much.

In olden days, the blinker lever would stay down to keep blinking. Turning the steering wheel back to straight after the turn would pop it back to off, or you'd just tap it up (past the detent) to turn it off. In Tesla's effort to eliminate buttons and keep things simple, it seems they not have found solutions which are quite as intuitive. I'm not suggesting the blinker lever should move itself, like the keys on an old player piano, but a lever with 5 positions the meanings of which change depending on circumstances seems unpolished.

Now that the blinker lever actually can steer the car, confusing behavior may not be a good thing.

 

[valem]

Is the rocking occurring independent of FSDb or is it from something like FSD wandering within the lane?

Here's a post about this issue.

Anyone else experiencing this side-to-side wobble? See video. It's pretty bad! It started with 11.3.6 and no change with recent update to 11.4.2

Only when the car is driving. I tried both FSDb or TACC same thing.
If I disengage I don't feel anything our of the ordinary. Even without touching the wheel the car goes straight as an arrow.

Below 50 mph hardly any wobble at all, but above 65 mph it gets really noticeable.
The faster I go and the more it wobbles. It's making my GF sick so this is serious!!

Only on AP/FSDb. Driving manual ... smooth as butter.

Had an alignment, made no difference.
Tires are in good shape; rear are new and front have lots of thread left... suspension inspected and all is tight ... no idea what to do

 

[Pdubs]

Not necessarily, it could be that you were already in Chill, and there's no setting to the left of that so it just stays in Chill.

Agreed. But like most people, I don't want to bring up the menus on top of the map or visualization.

So what I do now to avoid looking down there:
To be in Chill, left-click twice.
To be in Average, left-click twice then right-click once.
To be in Aggressive, right-click twice.

To set Minimal Lane Changes, you have to look down to hit the button after you manipulate the scroll wheel as above. Be quick or it will disappear and you'll have to start again.

I suppose that you may be able to program an S3XY button if you have those; I haven't checked into it. I also think the UI arrangement may change, as I haven't seen anyone praise how it works now.

If you have an accessory steering column display, there could be upcoming support depending on the manufacturer and model. I have the dual-round display and it was designed to show the standard Autopilot numeric follow-distance setting. It's now just stuck on "4" because that's the setting I had when I updated, and now FSDb is calling a different API routine for the new Chill|Average|Aggressive system.

So at some point, I will try to send in a support request for the display accessory, on the slim chance they will try to address it with a firmware update. But as of now, FSDb is a minority of the USA customer base, and the accessory display developers are in China where they don't even have FSDb yet. Maybe if the new settings become part of basic Autopilot, they'll realize they need to work on it.

What is this clicking and scrolling to change settings you’re talking about? I thought I had this all figured out.

 

[Pdubs]

First, I've never said FSD alone was safer than me because it's not. But when supervised by a good driver FSD I believe is safer.
As to my driving, I got my license in 1968 and have a perfect driving record. Not even a minor fender bender and never had a moving violation. And I drive in greater Boston not the easiest place. I drive somewhat faster than the average driver but never crazy fast.

@Ramphex- What's your driving record?

So if I’m reading you right, please correct me if I’m wrong, if FSD is safer with a good driver and a good driver is good without FSD, please share how FSD can make a good driver safer when the whole definition of a good driver is somebody with a record like yours who got that without FSD but somehow you’d be even better with FSD because you’re there to stoop it’s failures. Sorry. Im confused.

And before you ask, I’ve totaled 4 cars, been in 17 accidents and had my license suspended twice. The best accident I had was in I-75 in ‘82 when we were racing backwards. Trooper had clocked me at 87mph when a deer ran in front, er I guess behind me, got hung up in the rear drive and the car rolled over a couple times. That one was cool.

So I guess, grats on your good driving record. Not an easy feat up there. They have such bad drivers there, it’s actually a law that semi trailers have to have a sign on the back of their trailer that tells people they make wide turns. I mean really, what’s a better definition of, “Here’s your sign”?

 

[EVNow]

No sun here in Seattle!

We do get Sun in June, July, August and September.

 

[EVNow]

With the last few releases I've found FSDb to be inconsistent in taking right turns on red. It could be that these anecdotal reports are actually because of bugs in the right turn on red algorithms. Or it could be that they have a very buggy algorithm to recognize 'no right turn on red' and it's getting it wrong.

Could be just the map ? That would account for why FSD consistently ignores particular “no turn” and observes some other “no turn” signs.

 

[sleepydoc]

First, driving manually with blinkers not set to automatically turn off after a lane change. So I start the blinker, full press so it stays on, make my lane change and need to turn it off. It seems I need to press it in the same direction to stop it. True?

The default setting is similar to many cars: a 'half' press will blink 3 times, 'full' press will leave it on until you turn far enough or turn it off. To turn it off you can half press in either direction.

Second, now with it set to turn the blinker off after a lane change, sometimes it does not do turn itself off, perhaps after entering the freeway. But sometimes is does. Confusion and distraction occur when I turn it off manually, with a tap in the same direction, but I do that just as it turns itself off. So, my tap actually turns it back on, oops. Any suggestion on how to avoid this?

I'd say have a bit more patience and wait before turning it off to see if it's going to do it itself.

Finally, and here is where FSD comes in, the car is driving and decides it wants to change lanes. It gives me a chance to cancel. Should I tap the blinker, full tap the blinker, or tap in the other direction?

Half tap in the opposite direction. (I haven't tried anything else but this has worked for me.)

I know I could go out driving and experiment to figure out its logic, but freeways here are too fast and crowded to drive along randomly blinking, blinking. ;-) Actually, around here, FSD testers are the only one who do signal for lane changes, though I doubt other driver mind much.

I'd still suggest going out and playing on some less-busy roads or at a time when it's not as busy so you are familiar and comfortable with the blinker (and the rest of the car's operation).

In olden days, the blinker lever would stay down to keep blinking. Turning the steering wheel back to straight after the turn would pop it back to off, or you'd just tap it up (past the detent) to turn it off. In Tesla's effort to eliminate buttons and keep things simple, it seems they not have found solutions which are quite as intuitive. I'm not suggesting the blinker lever should move itself, like the keys on an old player piano, but a lever with 5 positions the meanings of which change depending on circumstances seems unpolished.

Part of the problem is Tesla needs the blinker to function in ways that the old mechanical levers did and could not.