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:
That was my impression too. Plus there's too much going on in that intersection for FSD to instantly slice though a traffic break (FSD path) with human like precise steering. FSD on my car has never worked anything like that. Most likely FSD alerted the driver to take over well after the danger passed. :)
Even a broken clock is right twice a day... you need to smoke hopium as much as I do....I am hanging out with Elon this week, so I have to be nice.
 
  • Funny
Reactions: bhakan and jebinc
Plus it was very unsafe and completely unnecessary to avoid the collision. Looks like pure human panic to me (albeit capably avoiding running directly into any stationary objects) - for sure the second movement, anyway.
If the driver was in a Subaru they would of been dead for sure. Slow, terrible brakes and handling....The street would have Subaru driver blood flowing like a river.
 
  • Informative
Reactions: AlanSubie4Life
Sure.

According to reports to the CA DMV Waymo has one disengagement per 17k miles driven (2022). In city driving Tesla has a disengagement ever 5-15 miles according to FSD Community tracker 2023. That's 1000-2000x.

If we only count critical* DE:s for Tesla they are at about 35 miles per DE, which would only be 500x worse than Waymo, but 1000x worse than the top reporters 2022.

So I'd say "three orders of magnitude" (1000x) is a pretty fair ballpark estimate all things considerered.

*) Avoid accident, taking red light/stop sign, wrong side of the road, unsafe action

Sources:
Do you legitimately not understand the strategic differences between the approach Tesla is taking, and the approach Waymo is taking?

How much does the hardware suite for Waymo’s AV cost?

How many times do remote operators intervene on Waymo vehicles?

Can Waymo drive anywhere in North America, or take you between any two large disconnected cities?

What is Waymo’s coefficient of drag at highway speeds?

Can I buy a Waymo vehicle?

If a new road is built or an old one removed, can Waymo handle that without remapping rhe area?

Surely you understand the difference. All of the above answers will make it quite clear that Waymo’s solution is not scalable. That’s ok, they have different goals—not knocking them. But they are apples and oranges to Tesla. Comparing them just doesn’t make sense.
 
Last edited:
Nightmare Before Christmas question: What will the Holliday update include?

  1. 11.4.7.1.1.1.1.1.1/11.4.4.1.1.1.1.1.1
  2. 11.5.x
  3. 11.69.x
  4. 11.420.x
  5. 12.x (the mythical)

Nightmare Before Christmas question: What will the Holliday update include?

  1. 11.4.7.1.1.1.1.1.1/11.4.4.1.1.1.1.1.1
  2. 11.5.x
  3. 11.69.x
  4. 11.420.x
  5. 12.x (the mythical)
11.24.23 will be released on Black Friday. You need to bring your car to camp overnight to get it. Nobody will have V12 without having this version pre-ordered.
 
Last edited:
Do you legitimately not understand the strategic differences between the approach Tesla is taking, and the approach Waymo is taking?
I do. Tesla is marketing a Level 2 system as capable of future autonomy with just a few software upgrades since 2016 and Waymo is deploying actual autonomous vehicles.
How much does the hardware suite for Waymo’s AV cost?
Not sure how this is relevant. They deem that hardware necessary to solve the problem they're working on. Do you think they are constrained by unit cost when they can have a car replace a human driver? They can always remove hardware down the line. Tesla cannot add hardware to my car or even upgrade the cameras/computer because of the cost involved even though they got 8k from me.
How many times do remote operators intervene on Waymo vehicles?
Once every 17k miles, as per my last post (DMV stats, 2022). Probably less often now.
Can Waymo drive anywhere in North America, or take you between any two large disconnected cities?
No they can't yet at the reliably level needed, and neither can Tesla or anyone else. Tesla can't even drive in a geofenced area of one square mile without supervision because it can't even stop at a red light reliably.
What is Waymo’s coefficient of drag at highway speeds?
What? LOL.
If a new road is built or an old one removed, can Waymo handle that without remapping rhe area?
Yes. HD-maps are used to improve safety. Think of it as a "future" sensor that lets them know about the likely look of an upcoming situation of let them see through occlusions (ex a blocked stop-sign).

If you can drive without HD-maps, you cannot deploy AV:s at scale. Waymo handle roadwork and repainting of road lines, et c just fine without operator interventions.

Surely you understand the difference. All of the above answers will make it quite clear that Waymo’s solution is not scalable.
Waymo drives better in any city that you drop them in at the first drive without maps than Tesla FSD. It's just that driving on that safety level is useless if you want to deploy robotaxi, so they don't.

I cannot believe people are so incredibly gullible that they still believe Tesla will become autonomous, after three years of amateur hour-like progress on their shitty hardware.
 
Last edited:
1. If Waymo has safety drivers intervening, they are not “actually autonomous”. And no, Tesla’s aren’t either.

2. Hardware cost is relevant for a number of reasons. Is the system scalable from a business perspective? Profitable? Replacement cost if it gets totalled in an accident?

3. CoD is important because your EV’s range drops to sh*t if you have to drive at highway speeds with tumors growing all over the outside of the car.

The whole point is that the approaches are vastly different and have vastly different scalabilities. Each has advantages and disadvantages. Tesla’s approach is much harder and will be slower to achieve reliability in basic stuff, but is scalable and will be incredibly more profitable at scale.
 
1. If Waymo has safety drivers intervening, they are not “actually autonomous”. And no, Tesla’s aren’t either.
Waymo don't use safety drivers. There can be no "interventions" in Level 4 or above, because there is no driver in the car and the system needs to complete the MRM by itself. The pedals and steering wheel are still mandated by federal law otherwise they wouldn't be there.
2. Hardware cost is relevant for a number of reasons. Is the system scalable from a business perspective? Profitable? Replacement cost if it gets totalled in an accident?
ROI on the sensors on a car is a few months in driverless. It's very unlikely that 100% would be damaged unless there is a fire, and then likely covered by insurance.
The whole point is that the approaches are vastly different and have vastly different scalabilities. Each has advantages and disadvantages. Tesla’s approach is much harder and will be slower to achieve reliability in basic stuff, but is scalable and will be incredibly more profitable at scale.
Imagine if Teslas could fly too after a software upgrade, or if Tesla had humanoid robots that actually was able to replace humans. Super profitable!

The main difference as I see it is that Tesla sells a a dream/hope of unsupervised autonomy without any real chance of getting there during the 3-5 coming years (on existing hw) and Waymo is deploying driverless at a 1000x better reliability, which was what started this exchange.
 
Last edited:
Here’s an interesting thing I noticed on the 11.4.7.3 build. my 2016 model s is still on original brakes after 129k miles. Since 11.4.1, Ive been wondering if I would burn through brakes faster since it seemed like FSD used the brake more than it needs to if it would just use regen sooner to slow down instead of barreling down the road and braking hard for lights, signs, and stopped vehicles. I assumed it was braking since the physical petal would depress. lately I noticed it still slows down hard and late, but I don’t see the pedal moving. i know Tesla can change the strength in software if they choose, but is it possible the vehicle on FSD is dynamically adjusting this? I see the regen bar spike to green and the pedal isn’t moving, but it feels stronger than regen when I drive one petal.
 
  • Informative
Reactions: FSDtester#1
i know Tesla can change the strength in software if they choose, but is it possible the vehicle on FSD is dynamically adjusting this? I see the regen bar spike to green and the pedal isn’t moving, but it feels stronger than regen when I drive one petal.
I use Scan my Tesla. One of the most interesting parameters I choose to display is Max Regen. This is a moment-by-moment number which for both my 2022 model Y LR and my 2023 model 3 RWD has a cap at 85 kW. That is A LOT. Trust me, if it used all of that in local street sub-35 mph driving you'd think it had gone berserk and was trying to put you through the window.

I believe that if FSD sees the need to get you slowed down, it won't limit regen use to what you'd get as a standard single-pedal foot up in the air number.

On the other hand, depending on the three main governing variables of State-of-Charge, Battery temperature, and recent power flow (in and out of the battery) the available max Regen is often less than 85. On a chilly day with a freshly charged battery I started with about 45 today, and saw as low as about 15. On other days when the first thing I did was to go down a 1000 feet of elevation in four miles, I've seen as low as 3.5 kW. That won't get you even grandmother-grade stopping.

When I'm in FSDb, which is nearly all the time, I routinely look ahead for reasons the car will need to slow anyway, and use the right scroll wheel to dial down the current max speed progressively before FSDb would have started slowing. I'm pretty sure I am avoiding some brake wear this way, and even saving a little energy consumption. I also just like the drive better that way.
 
I recently test drove a Model S that had FSD enabled and for some reason it got me to resubscribe to it on my Model Y (plus, my wife loves the FSD visualization -- perfectly reasonable reason to spend $200 a month). Anyway, I've been using it a little more and actually have enjoyed my time with it. I'm using it more as an advanced driver's assist rather than expecting it to really drive on its own, but it's made some boring stretches of my drives more interesting somehow. It also keeps me from speeding as much and I'm more patient with slow traffic.

Anyway, I've found that keeping one hand on the lower left of the steering wheel almost perfect for keeping the nags away as I tend to apply the needed torque without even trying and it's an easy position to apply torque if the dreaded blue flash does happen. I got the idea by watching a youtuber use his leg to hit the steering wheel. I still feel like the car is handling the driving, but I have my hand ready to intervene at any moment. Better than constantly moving the scroll wheel for me, anyway.

When traffic allows I let it handle some more advanced maneuvers and I'm pretty shocked that it's doing fine on the two roundabouts on my route to work. When I first got the car it would completely freeze in them. I'm only doing this when there is no nearby traffic as I have zero confidence that it would handle it with cars around me.

For all its many faults FSD is still pretty trippy when it does go long stretches driving itself. I just wish it would stop trying to change lanes all the bloody time.
 
And the basis for this feeling is ... ?

I'm assuming you are just like many of us here - non-experts. Right ? When even self-driving car experts don't know how to solve the problem, you & I surely don't know either.

you are correct I have never brought a full self driving car to market. since there is no such thing as self driving, where can you find a self driving expert?
I am an electrical engineer with nearly 40 years of experience in designing engineering and commissioning automation and controls. currently working for an automation and intelligence company, with a heavy push on machine learning. I have had many successes and many failures in my career. So many Automation strategies look great on PowerPoint, but the great ideas were hampered by unreliable sensors poorly sized equipment that had many control gaps, that you could not get to the setpoint, even if you knew what it should be. The main difference, is in the real world of Automation, every one of my projects, have penalties and liquidated damages for lack of performance, or not supplying the product contracted for..
I am saying my opinion is that Elon’s approach has failed. He has not delivered the product, sold me in over five years. do any of you actually believe that even by 2025, a 2018, Tesla will be operating in full self driving without operator 100% responsibility and supervision, and that vehicle will be approved by regulators, and you will be able to insure that vehicle. The lie is getting very old, of course everyone knew it was hard, the smartest guy in the room just claimed to have figured it out I believed him for a while, but it is clearly BS by anyway, you could measure it.

But it’s sad that long-term owner, an automation professional, continues to be questioned, by Tesla stock shills.
I thought this was an owners, opinions and viewpoint site, but I guess I got to spend more time doing the Elon Circle flirt.
IMG_1704.jpeg
 
I am saying my opinion is that Elon’s approach has failed. He has not delivered the product, sold me in over five years. do any of you actually believe that even by 2025, a 2018, Tesla will be operating in full self driving without operator 100% responsibility and supervision, and that vehicle will be approved by regulators, and you will be able to insure that vehicle.
I honestly feel bad for the early adopters (pre-2019) - I think you guys have a case to have Tesla refund you the FSD money. Post 2019, I don't feel bad at all. Nowhere in my ordering experience did Tesla promise me my car would be L4 or L5, or mention robotaxi. I read everything indicating it is an L2 driver assist feature where I pay attention and the car assists me with driving. The only place I heard people talking about L4 or L5 was on Twitter with Elon's statements. When I bought my car I was not on Twitter, don't care about Twitter, and didn't know anything about what Elon said about anything. I just read the safety government crash-testing data on the car, Edmonds and Car and Driver reports about the safety, reliability, and brand loyalty for Tesla. I learned about the build-quality issues with the Model Y, and what to look for when inspecting the car before accepting delivery.

Yet another reason I think social media sucks - if people stopped listening to Elon on social media, and just paid attention to what is being said on the website during the ordering process, and what disclaimers they have to accept when they go through the agreements, some of these arguments wouldn't be a big deal.
 
I honestly feel bad for the early adopters (pre-2019) - I think you guys have a case to have Tesla refund you the FSD money. Post 2019, I don't feel bad at all. Nowhere in my ordering experience did Tesla promise me my car would be L4 or L5, or mention robotaxi. I read everything indicating it is an L2 driver assist feature where I pay attention and the car assists me with driving. The only place I heard people talking about L4 or L5 was on Twitter with Elon's statements. When I bought my car I was not on Twitter, don't care about Twitter, and didn't know anything about what Elon said about anything. I just read the safety government crash-testing data on the car, Edmonds and Car and Driver reports about the safety, reliability, and brand loyalty for Tesla. I learned about the build-quality issues with the Model Y, and what to look for when inspecting the car before accepting delivery.

Yet another reason I think social media sucks - if people stopped listening to Elon on social media, and just paid attention to what is being said on the website during the ordering process, and what disclaimers they have to accept when they go through the agreements, some of these arguments wouldn't be a big deal.
So, a few weeks ago, here I was in my Tesla Model 3, 9/2018 vintage, RWD, LR. Tesla offered a deal to transfer the FSD to a new Tesla, so long as said Tesla was bought before 9/30/2023.

I bit. And, now I'm driving a slightly different Tesla, with FSD. M3, LR, but AWD now. Just turned the odometer over on 1500 miles after a trip from NJ to Boston and back.

By the way: 11.4.4 works fine. There were the usual (rare) disengagements; no problems with the windshield wipers, either on the way up (it was sunny) or the way back (rain/drizzle all the way). As 11.4.4 does, it does have issues with short on-ramps and off-ramps; but no running red lights with this guy. But the highway miles were excellent. On the Merritt, in particular, it would move in and out of the passing lanes sans problems. I at least partly blame the 2023 map update for some of the better behavior.

It actually did an pretty decent job getting onto the George Washington Bridge westbound from the Hudson parkway; in particular, in the three places where two lanes take off that-away, it properly placed itself in the correct lane (usually, not the restricted lane that had to turn), making the trip through there Not Bad on a Monday night.

As I've occasionally stated around here, I don't seem to have the problems that many of you appear to have with FSD. Maybe it's something about driving around in California or just Out West of here. But the squiggly, we-don't-believe-in-right-angle-roads-we're-colonials intersections doesn't seem to give FSD the fits. Either before the new car (on 11.4.7.2) or with it (on 11.4.4).

And, as you all must have noted, even with the trade-in, the $7500 fed tax credit, and NJ's contribution, I've put my money where my mouth is: On FSD.

Back to feeding the trick-or-treaters. There's a lot of dinosaurs out front these days.
 
On 11.4.4 (HW4)
The one consistent improvement I've had over the past few months is highway merges. They used to be terrible. Now except when highway traffic is stop and go and FSD wants to fly by everyone merges have been decent. Blinker gets used and most of the time FSD moves over before the end of the zip lane. Other than this change FSD is the same for me. Still a schizophrenic child.
 
Twice now 11.4.4 has, at this intersection, where the programmed route is to turn West/left at the light when coming from the South (as pictured), turned on a right turn signal and entered the right turn lane. I have to take over, take it out of the right turn lane and cross the center lane into the left turn lane.

In no world does this make sense to me. Although the intersection was recently revised, adding a right turn lane, there's no reason for the car to choose to go into the right turn lane when the route clearly indicates pulling into the left turn lane, which has not changed for decades.

This is on the main north/south route through the state, not some backroad.

This right turn when a left turn is the route has also happened at another intersection, without a light.

Anyway, I am done with FSDb for a while. It really does strange stuff here. I hope it improves.

1698792945169.png
 
  • Informative
Reactions: HitchHiker71