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:
Still waiting for the apologists to make up excuses for the Junk’s behavior in the 3 videos I posted today.
I haven't spent anytime today watching videos on this thread. Just went thru today's posts and don't see any videos you posted so I could watch them.
Did I just miss them or where you referring to your comments about other peoples posted videos?
 
At first when I got 11.4.2 it was very bad but after driving it for about a week or so it improved and my interventions have been much less mostly accelerator pedal pushes, but my disengagement have gone way down. I was really pissed at first but now it’s starting to get better. Not sure why maybe the NN
My experience also, although I did had some regressions again, about a week ago.

The general understanding is that the NN doesn't learn within a version and not in the car itself. However, the system is Increasingly taking advantage of downloaded map and route-specific information that comes from the server when you navigate a new route, as well as mid-route sometimes. Updates to that server database are responsible for any actual improvements (or regressions) that occur within a release timespan.

As has been pointed out also, there is the issue of learning and familiarization on the human user side. This can affect the interpretation of otherwise somewhat random variations in the FSD performance, after a release has been on the car for a few drives.

All this is discussed a few pages back, prior to some pages of the usual ranting and bickering.
 
Last edited:
That whole metric is idiotic. People with FSD enabled are on the whole going to be far more attentive than those who don’t have it enabled, because many (most?) expect it to royally mess up at any time. It’s just smoke and mirrors from Tesla. If I left FSD enabled and never took over I’d have had any number of accidents already. My car would not be road worthy at this point.

What Tesla should be saying is that drivers with FSD enabled are safer than drivers without, because drivers live in a permanent state of hyper awareness with FSD enabled. You take the average driver of any other car versus a Tesla with FSD enabled and the driver unable to disable FSD, the human driver will on average be far safer. FSD in of itself, today, does not make cars safer. But you certainly don’t see Tesla stating that the constant fear FSD drivers live in is contributing to safety.

Now, when it comes to AutoPilot, that’s a different matter. But FSD on e.g. Seattle city streets? Get outta here. I’ll put my money where my mouth is, too, in any way you like.

Now you’ll just respond: See! FSD = safer human drivers!

I’ll roll my eyes at that. You remove seatbelts from cars and you’ll have slower/safer drivers, too. 🙄
I want to reply to this with my own experience recently. I drove my 2020 MX around in Orange County CA, mostly on 11.3.6, and had it engaged at least 90% of the time because it made driving in the California traffic much easier. Sure there were occasional errors, but the overall stress of driving was greatly reduced for me, quite the opposite of what you postulate. Driving up to Oregon was done almost exclusively with FSD engaged. I had no interest in disengaging it for the most part. Once arriving in Oregon, I drove back and forth between the I5 corridor and US 101, plus up and down the 101, through towns etc., considerably more relaxed than driving it manually. Now I have traded in my 2020 for a new MX with HW4. The result of this is that I do not have FSD yet, and I am very unhappy about that. I drive around highway and city with only Autopilot, and that requires much more attention and intervention from me than FSD. Autopilot is comparatively much stupider than FSD, and much less capable. I now have to adjust the speed up and down all of the time, disengage for speed changes for turns, and it is useless in town. I feel much less safe without FSD, quite contrary to what you are saying. Of course, your comment about cars without seatbelts being safer than with is ludicrous.
 
I want to reply to this with my own experience recently. I drove my 2020 MX around in Orange County CA, mostly on 11.3.6, and had it engaged at least 90% of the time because it made driving in the California traffic much easier. Sure there were occasional errors, but the overall stress of driving was greatly reduced for me, quite the opposite of what you postulate. Driving up to Oregon was done almost exclusively with FSD engaged. I had no interest in disengaging it for the most part. Once arriving in Oregon, I drove back and forth between the I5 corridor and US 101, plus up and down the 101, through towns etc., considerably more relaxed than driving it manually. Now I have traded in my 2020 for a new MX with HW4. The result of this is that I do not have FSD yet, and I am very unhappy about that. I drive around highway and city with only Autopilot, and that requires much more attention and intervention from me than FSD. Autopilot is comparatively much stupider than FSD, and much less capable. I now have to adjust the speed up and down all of the time, disengage for speed changes for turns, and it is useless in town. I feel much less safe without FSD, quite contrary to what you are saying. Of course, your comment about cars without seatbelts being safer than with is ludicrous.
Mostly highway driving?
 
Exactly. If there's traffic and you're doing 40 or less, you are legally free to read a book, watch a movie, mess with your phone, etc because the car- not the human- is driving.

If traffic clears up the car does not just keep doing 40.... It alerts the driver they need to take over in the next 10 seconds, then when it hits 40 it KEEPS accelerating just like autopilot would to whatever the relevant highway speed is, but now it's just L2 ADAS and the human is driving.




The difference is Mercedes believes their system is good enough to, under these limited circumstances, take responsibility for the entire driving task including liability. Nobody else, Tesla includes, appears to believe that about their own systems.
Personally I believe Tesla's AP is good enough to do what MB is doing. In the 3 years I've had my MY I've never had AP mess up. FSD is another story, but it's another animal, too.
 
  • Like
Reactions: GSP and Yelobird
That whole metric is idiotic. People with FSD enabled are on the whole going to be far more attentive than those who don’t have it enabled, because many (most?) expect it to royally mess up at any time. It’s just smoke and mirrors from Tesla. If I left FSD enabled and never took over I’d have had any number of accidents already. My car would not be road worthy at this point.

What Tesla should be saying is that drivers with FSD enabled are safer than drivers without, because drivers live in a permanent state of hyper awareness with FSD enabled. You take the average driver of any other car versus a Tesla with FSD enabled and the driver unable to disable FSD, the human driver will on average be far safer. FSD in of itself, today, does not make cars safer. But you certainly don’t see Tesla stating that the constant fear FSD drivers live in is contributing to safety.

Now, when it comes to AutoPilot, that’s a different matter. But FSD on e.g. Seattle city streets? Get outta here. I’ll put my money where my mouth is, too, in any way you like.

Now you’ll just respond: See! FSD = safer human drivers!

I’ll roll my eyes at that. You remove seatbelts from cars and you’ll have slower/safer drivers, too. 🙄
It's comedy for sure. As many have already stated a sizeable # of FSDers disengage FSDj for riskier scenarios and let it ride when there's less risk. Marketers talk from both sides of the mouth of FSD being safer than a human driver while also warning FSD will do the wrong thing at the worst time and there will be unknown issues and drivers need to be paranoid. It's all about the driver.
 
The reason for that is I strongly dislike the car's tendency to drive up to a light at full speed and then brake firmly. Yes, it stops, but it doesn't drive the way I drive.
I share your distaste for FSDb's braking profile. It at least has improved a bit with the more recent releases. It will still keep accelerating even when the light is red, though. I prefer to do as you - adjust my speed so I hit the light (or line of cars) right as the light changes and I never actually have to stop.
 
  • Like
Reactions: GSP
That whole metric is idiotic.

So you insist there's no evidence the metric has been met- then when presented with evidence you call the metric idiotic.

There's an untrue Scotsman looking for you.


Get outta here. I’ll put my money where my mouth is, too, in any way you like.

Except, apparently, actual accident rates.

Because that measure objectively proves you wrong, so it's "idiotic"

I agree there's an apologist thing going on, but it ain't from me :)



Personally I believe Tesla's AP is good enough to do what MB is doing.

Apparently Tesla does not share your belief.
 
A couple good long drives today on 11.4.2. Went from my home in the D/FW area to the Henryetta, TX supercharger and back. This 135 mile route takes US287, which gave me no end of grief last year on NOA. It's a four lane divided highway with surface grade crossings. There's lots of opportunity for heat reflections to create mirages. Last year, I NOA produced a bumper crop of PB events and multiple Red Hands of Death on this road.

Today was an entirely different story. My route included travelling north on I-35 from Dallas to Denton, then west on US380 to US287, and then up to Henryetta. This is all highway driving.

I had one or two minor PB events, though none on US287. I did one disengagement when FSDb could not figure out how to transition from US380 to US287 with most of the turn lane coned off for construction. I also used the turn stalk a few times to change lanes, as sometimes FSDb does not like to move out of the left lane on US highways.

On the return trip, I had a zero intervention drive the entire way. No disengagements. No manual lane changes. And no accelerator usage. Just pure FSDb for 135 miles. Nice.

Special mention should go for navigating I-35 traffic with no drama. FSDb is worth the money just to avoid driving this abomination of a road by hand!

I expect FSDb to screw up royally tomorrow to keep it's average down. But today was great.
 
Let’s entertain some of the “anecdotal” things mentioned earlier. How about Autopark (Beta)? How long has that been a thing?

While getting my industrial strength hemorrhoid cream today, I recorded a video of Autopark in action. Let’s watch how well it performs on a “flagship refresh Model S”


A good ol 60 seconds to back into a spot. Dang it’s good! Revolutionary. Puts into perspective what to expect from the rest of the beta Junk.

Here’s a couple more vids. Turn Signals (Beta)


I haven't spent anytime today watching videos on this thread. Just went thru today's posts and don't see any videos you posted so I could watch them.
Did I just miss them or where you referring to your comments about other peoples posted videos?
All of those iCloud links are videos…
 
  • Helpful
Reactions: jebinc
I feel much less safe without FSD, quite contrary to what you are saying.
2 days back just getting out of my neighborhood I was taking ULT. There was a car blocking the view to the left partially. Just as I was pulling into the intersection a car came out from behind the parked car. I had to slam the brakes.

If I had engaged FSD it would have probably been safer ;)

I had not yet put in the destination. .. so was driving manually.
 
So just so everyone is on the same page for this discussion:

When there are videos posted of 0 intervention or 0 disengagement drives, those are cherry picked or edited to hide errors.

When there are videos posted showing errors, those are not cherry picked or edited to hide successes.

We should extrapolate error videos to show that FSD Beta is a huge, dangerous failure.

We should not extrapolate 0 disengagement videos to show that FSD Beta is successful.

Am I getting this right?