Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

FSD rewrite will go out on Oct 20 to limited beta

This site may earn commission on affiliate links.
It's crazy I even think this, but 6-9 months lol. We've seen essentially every human maneuver from this beta, even unintuitive or difficult to program ones like:

Where is my robot taxi @powertoold
Its almost 6 months. You said it will reach human level in 6 months.
So where is it? Others back me up that that's what you said.

That math is bit off.
Tesla 1/3 improvement every month in interventions
My assumptions are: start at 1 intervention every 5 miles. Every month improve by 1/3rd. 37 months before reach 150K miles intervention.

I'm willing to take a bet on that. :) Tesla will not release level 3 or 4 wide release for cities in that time frame or that published city intervention rate is more than 150K miles by that time frame.

If you start off with 1 intervention every 7 miles and every two weeks the number reduces by 1/3, than after 6 updates you are at one intervention every 156,865 miles...

Basically, it looks like Tesla has won the race to a widely deployed FSD system. Just my insight.

You said game over, Elon said game over 2 years ago. But where is it?
 
Interesting thought, BUT couldn't the car just not allow FSD during those snow conditions?

Sure if they want to finally admit the cars sensors are incapable of L5 self driving like anyone who has owned one for years outside of Southern California already knows.

They seem...reluctant...to admit that though.
 
Sure if they want to finally admit the cars sensors are incapable of L5 self driving like anyone who has owned one for years outside of Southern California already knows.

They seem...reluctant...to admit that though.

We do quite well (well that's debatable) at L5 driving, with much less sensitive, and poorer/faulty sensors than the car already has.
I know I don't have radar which can see through fog/rain/snow, under other cars, and have eyes in the back of my head, and in my temples.
My ears also aren't very good at sonar range finding.

My brain, makes up for the deficiencies in my senses.
If I had better senses, I wouldn't need so much brain-power to make up for the poor sensor set.

How does a person drive, when they can't see the road markings through the snow? Follow the leader, Follow the tracks, look at the other markers which indicate where the drivable area is. No doubt, the car can't do that yet. We also get it very wrong, much of the time in the snow, and probably just get away with it the rest of the time.
 
We do quite well (well that's debatable) at L5 driving, with much less sensitive, and poorer/faulty sensors than the car already has.

Not really.

For example your eyes aren't covered in mud or slush in bad weather. Since they're inside the car.

The cameras-not so much.

Your eyes also don't fog up like the B-pillar cameras do in some weather.

Also your neck and body can move if you need better visibility around a corner than your normal seating position- the car cameras can't (which is why some FSD beta videos show the car nearly pulling into traffic to get a better view when making turns and still having issues)



I know I don't have radar which can see through fog/rain/snow, under other cars, and have eyes in the back of my head, and in my temples.
My ears also aren't very good at sonar range finding.

My brain, makes up for the deficiencies in my senses.
If I had better senses, I wouldn't need so much brain-power to make up for the poor sensor set.


Yes- when the car sensors are working well, the car absolutely should have better objective observational abilities than you in many, but not all, cases... (your eyes have much longer range and better resolution than the cameras as a contrast)

But the sensors are often not working well in bad weather and have no way to compensate for it- and no redundancy if even one camera on either side is obscured.


How does a person drive, when they can't see the road markings through the snow? Follow the leader, Follow the tracks, look at the other markers which indicate where the drivable area is. No doubt, the car can't do that yet. We also get it very wrong, much of the time in the snow, and probably just get away with it the rest of the time.


Ah, but L5 can't just "hope it gets away with it"

Too much liability.


Right now even just NoA, which is a long long long way from L5, turns itself off in even moderate rain because the sensors can't see well enough.

In BAD rain even basic AP turns itself off sometimes.


The current sensor suite could ABSOLUTELY be good enough for L3 on the highway (which in my personal case is enough to satisfy me with the $ I spent on FSD, since that'd cover 95% of my driving and I could read a book while "driving)

It MIGHT be good enough for L4, with one of the restrictions on ODD being "No worse than light or moderate bad weather" and the car pulls itself over and parks if the weather worsens (since L4 there's no human required to take over).


If you want me to buy that it has any shot at L5 you're gonna need to add some more sensors.

The suggestions I've made in this regard are:


Rear radar (which would also enable the cross traffic alert folks always complain the car lacks)

Side-facing cameras on the front of the car (to see sideways on intersections without needing to pull partly INTO the intersection as the B-pillar cameras occasionally need too.... and also providing redundancy to the sides)

I think that'd be a minimum...

In an ideal world there'd also be a second pair of redundant rear/side facing fender cameras maybe on the rear fenders to back up the ones on the front fenders.... and maybe a couple of low-mounted fisheye cameras to offer a legit 360 parking view the car lacks HW to offer today.


(note at no point do I suggest Lidar's needed- I don't think it is- but the current suite ain't getting you better than L4 with weather restrictions)
 
Not really.

[cut]

Ah, but L5 can't just "hope it gets away with it"

The current sensor suite could ABSOLUTELY be good enough for L3 on the highway (which in my personal case is enough to satisfy me with the $ I spent on FSD, since that'd cover 95% of my driving and I could read a book while "driving)

It MIGHT be good enough for L4, with one of the restrictions on ODD being "No worse than light or moderate bad weather" and the car pulls itself over and parks if the weather worsens (since L4 there's no human required to take over).


If you want me to buy that it has any shot at L5 you're gonna need to add some more sensors.

The suggestions I've made in this regard are:


Rear radar (which would also enable the cross traffic alert folks always complain the car lacks)

Side-facing cameras on the front of the car (to see sideways on intersections without needing to pull partly INTO the intersection as the B-pillar cameras occasionally need too.... and also providing redundancy to the sides)

I think that'd be a minimum...

In an ideal world there'd also be a second pair of redundant rear/side facing fender cameras maybe on the rear fenders to back up the ones on the front fenders.... and maybe a couple of low-mounted fisheye cameras to offer a legit 360 parking view the car lacks HW to offer today.

(note at no point do I suggest Lidar's needed- I don't think it is- but the current suite ain't getting you better than L4 with weather restrictions)

I guess we will see. Don't think anybody knows yet, but I think the existing set will get us a long way there. Those extra sensors would definitely give better than human visibility, and wouldn't argue against them.

The move to 4D modelling though, like we as humans do, should offset the weather related problems. We as humans, also have similar problems with the screens windows being obscured by rain/snow etc.

Having a model of the world, means that you can predict behavior of the world, even if you can't see it every frame. You know there is a car beside you, because you saw it 10-15 seconds ago in your mirrors approaching you, even if you can't actually see it right now.

I agree, the $ I spend on FSD has already been a good buy, even with less than L3 capability. My commute is 30miles of freeway and 5miles of side-streets at either end, often in heavy/slow traffic. Can't read a book/switch off quite yet, but being able to concentrate on managing the car/journey, rather than directly controlling speed/steering every instant make it worth it.

TBH. I think FSD will end up being L3+ with some aspects of L4. Would be quite happy if FSD ended up being level4 capability on freeways, under 'reasonable' driving conditions, and L3 on side roads. If it can't manage heavy rain/snow in L4 mode, to give option of dropping to L3, or abort/pull off freeway at exit or onto shoulder.

Don't think L5 really applies much, and probably wouldn't want to pay for it. e.g. Let's take the Cybertruck, fully automated, rock-crawling. I'll just sit at home, and watch the video while the truck goes off and does it's thing. (extreme example)
 
  • Like
Reactions: mikes_fsd
I guess we will see. Don't think anybody knows yet, but I think the existing set will get us a long way there. Those extra sensors would definitely give better than human visibility, and wouldn't argue against them.

The move to 4D modelling though, like we as humans do, should offset the weather related problems. We as humans, also have similar problems with the screens windows being obscured by rain/snow etc.

Except, again, humans eyes are inside the car, not getting mud and snow and slush piled on them.

And we have wipers for the windshield those eyes see through.

The car- apart front the front cameras somewhat- has no way to clear an obscured camera, and nothing keeping it from getting obscured in bad weather-- and again except for the front cameras, 0 redundancy to handle a single camera being blocked.




Having a model of the world, means that you can predict behavior of the world, even if you can't see it every frame. You know there is a car beside you, because you saw it 10-15 seconds ago in your mirrors approaching you, even if you can't actually see it right now.

It can't really work like that in most cases.

4D is useful so that it understands, with all the sensors clear, the car it saw in the front camera is now the same car is sees in the side camera as you are passing it....

But If my side camera is blocked the car doesn't know if it's safe to change lanes in that direction.

Because the car has no way to know if someone else changed lanes to be beside me (or if someone who was beside me changed lanes to not be) for example.

It's never going to move the car into a space the sensors aren't letting it see based on where it "thinks" a known-to-be-moving thing ought to be based on where it was earlier when it could see it.

A human might choose to change lanes without looking (either in a mirror or over their shoulder) but no car maker is ever going to take that chance.



I agree, the $ I spend on FSD has already been a good buy, even with less than L3 capability. My commute is 30miles of freeway and 5miles of side-streets at either end, often in heavy/slow traffic. Can't read a book/switch off quite yet, but being able to concentrate on managing the car/journey, rather than directly controlling speed/steering every instant make it worth it.

That's very very similar to my own drive.

That said- the $5000 I paid for EAP already covers all the existing functionality for the highway part.

So I'm gonna need at least L3 to let me read a book to have gotten any value out of the extra 3k for FSD.


TBH. I think FSD will end up being L3+ with some aspects of L4. Would be quite happy if FSD ended up being level4 capability on freeways, under 'reasonable' driving conditions, and L3 on side roads. If it can't manage heavy rain/snow in L4 mode, to give option of dropping to L3, or abort/pull off freeway at exit or onto shoulder.

Yup, that's the best I think we can expect with current sensor suite... (and I'm still somewhat dubious of getting above L2 in city driving... there just seem too many things that can happen too quickly in such environments to allow a time delay before a human has to take over)
 
Sure if they want to finally admit the cars sensors are incapable of L5 self driving like anyone who has owned one for years outside of Southern California already knows.

They seem...reluctant...to admit that though.

The first place to admit that they might be remotely off in their prediction is in the filed 10-K. It is on their investor’s page. In this they give an outline of all the risks that might come up doing business. If you go through it you will find all kind of outlandish risks stated, just to cover themselves against possible future litigation. What struck me is that there is no provision whatsoever about not being able to have FSD and implementing a Robotaxi network. Tesla is still very sure about the outcome.

tsla-10k_20201231.htm
 
The first place to admit that they might be remotely off in their prediction is in the filed 10-K. It is on their investor’s page. In this they give an outline of all the risks that might come up doing business. If you go through it you will find all kind of outlandish risks stated, just to cover themselves against possible future litigation. What struck me is that there is no provision whatsoever about not being able to have FSD and implementing a Robotaxi network. Tesla is still very sure about the outcome.

tsla-10k_20201231.htm
Tesla will deliver city NOA and consider that end of their responsibility w.r.t. FSD money.

That’s why there is no risk listed in the statement. Companies are very careful about admitting responsibility for anything in their financial statements. Those will be (and have been) used in litigation.
 
But that L3 letting you read a book is a lot more expensive than $3k. How much does years of Uber cost ?

That $3k was the discounted value based on risk.


...what?

3k is what I paid to go EAP->FSD.

So that's how much it'd cost me if it gets delivered.



Tesla will deliver city NOA and consider that end of their responsibility w.r.t. FSD money.
.


Not mine.

Not anybody who bought before roughly March 2019.

All us folks were promised a lot more than that- at minimum L4 with a pretty broad ODD- arguably L5.

The post 3/19 buyers yeah that FSD $ can be recognized fully as they got what was promised during sale.
 
...what?

3k is what I paid to go EAP->FSD.

So that's how much it'd cost me if it gets delivered.






Not mine.

Not anybody who bought before roughly March 2019.

All us folks were promised a lot more than that- at minimum L4 with a pretty broad ODD- arguably L5.

The post 3/19 buyers yeah that FSD $ can be recognized fully as they got what was promised during sale.

this is an important thing that a lot of folks don't get. The way FSD is sold now vs when we bought it is quite different and is going to be a lot harder for tesla to satisfy. It's likely to be pretty interesting when people who replaced their 2016/2017/2018 FSD-licensed cars with newer ones realize they paid more money for fewer commitments from tesla.
 
  • Disagree
Reactions: mikes_fsd
Sorry to ask what is probably a total newb question, but I didn't see a sticky post or anything... how exactly is it determined who gets FSD Beta? Does anyone even know, outside of Tesla? I got my M3P with FSD in October and was really excited when I first heard about the beta and started seeing Youtube videos. IMO, it seems FSD in it's current state on my car is borderline next to useless. I hate to say it, but I likely would not have paid for it if I'd tested it, first. I like auto-change on the highway, but it sometimes is too slow to even make that worthwhile. Meanwhile I'm seeing some Youtube videos from channels like AI Driver and the Beta version looks to be many factors better. How does one actually get it, though?
 
Sorry to ask what is probably a total newb question, but I didn't see a sticky post or anything... how exactly is it determined who gets FSD Beta? Does anyone even know, outside of Tesla? I got my M3P with FSD in October and was really excited when I first heard about the beta and started seeing Youtube videos. IMO, it seems FSD in it's current state on my car is borderline next to useless. I hate to say it, but I likely would not have paid for it if I'd tested it, first. I like auto-change on the highway, but it sometimes is too slow to even make that worthwhile. Meanwhile I'm seeing some Youtube videos from channels like AI Driver and the Beta version looks to be many factors better. How does one actually get it, though?

Tesla secretly invites people to get FSD Beta. It is not something you can apply for. Tesla has to pick you. Tesla did say that Tesla owners who have a proven safety record are eligible for FSD Beta but I am sure there is more to it than that. I would guess that Tesla might also pick owners based on location if they need certain type of data. I also suspect that Tesla picks influential youtubers to get FSD Beta because Elon wants videos of FSD Beta to help market and promote Tesla. It's free viral marketing for Tesla. But beyond that, I don't think we know the exact process that Tesla uses to select FSD Beta users.
 
  • Helpful
Reactions: powertoold
this is an important thing that a lot of folks don't get. The way FSD is sold now vs when we bought it is quite different and is going to be a lot harder for tesla to satisfy. It's likely to be pretty interesting when people who replaced their 2016/2017/2018 FSD-licensed cars with newer ones realize they paid more money for fewer commitments from tesla.

The way I remember it, when I bought it with the car, is that FSD was being 'advertised' or sold as, being capable of being added to the Robo-Taxi fleet, (subject to regulatory approval, and launch of the Robo-Taxi service, and development risk).

That was the 'value add' or potential upside of the risk. That the value of the FSD system being bought into at risk would likely/possibly give a return in the value of the car, as the value would go up in response to the (future) development and capability.

So expecting the FSD feature to be at-least as capable as what's needed for the Robo-Taxi service, IF it ever comes to fruition, and I still have the car.

Robo-taxi, by definition needs to be L4. "Unsupervised, under most, but constrained, conditions". Not necessarily L5, all conditions.

That was part of the risk of buying upfront Beta/Unfinished
 
The way I remember it, when I bought it with the car, is that FSD was being 'advertised' or sold as, being capable of being added to the Robo-Taxi fleet, (subject to regulatory approval, and launch of the Robo-Taxi service, and development risk).

The language on the configuration page has changed, but as you say, the language on this page remains largely unchanged: Autopilot
 
  • Like
Reactions: mikes_fsd