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.
L5 is a non-starter. Forget some edge case - can they handle normal traffic in places like India ? How about hairpin bends in the Himalayas ?

I think we should only talk about L4.
Should Tesla try solving India driving or should Indian Govt standardize to better traffic rules?
L5 on US can happen without needing to solve all the extreme cases in ROW. cheers!!
 
  • Like
Reactions: Electroman
FWIW I had more disengagements on 12.3 yesterday than I usually had on 11.x.... in particular on 3-4 lane each way (but non-highway) roads it was constantly make weird lane change choices and bailing out halfway between slow lane changes. I'd seen this complaint in the FSD threads in recent days but first time on roads where I ran into it myself.

On small 1-2 lane each way roads its behaved a lot better- noticeably so-- but having seen a decent bit of both good and bad in it now nothing makes me think we're any nearer to robotaxis or anything like that (ie still years away and probably never on the HW3 and older cars)
 
Timestamp to where he had to turn? I didn't see FSD on, the car was just sitting there displaying FSD path prediction in gray, not in blue.

And we are talking about right turn where side camera is not good enough, not going straight where b-pillar is not enough.


I mean- we're talking about both.... and in both cases insufficient B-pillar view on obscured intersections---- turning right or going straight BOTH require being sure there's no oncoming uncontrolled traffic from your left side.

Your "side camera" idea continues to be nonsensical, the fender cams look straight back, you can't angle the car that much at a turn that's obscured to the B-pillar to be useful without either already being in the flow of cross traffic or being so far angled you're blocking 2 lanes doing it.


The new Autopark in action. Hopefully V12 merges with the main branch soon so we can all test this out:



So... still just slowly backing into spaces....? That's the same autopark USS cars have had for years now, just with a bit better visualization of potential spots.


I mean I guess kudos for finally getting non-USS PART way to parity with USS ones (still missing summon right?) but it sure took em long enough.
 
I think this misses the point.

If the car is say 5x better at lane changing because it can look multiple directions at once- but is 5x worse at going through visually obstructed intersections safely because the B-pillar angle sucks- that's not going to be an acceptable robotaxi.

The system will need to do ANY of the DDT at minimum no worse than the average human.... and do MANY (but not necessarily all) parts of it better, before it'll be generally acceptable to take the human out of the loop.
Not really, it might miss YOUR point, but mine was addressing claims about "the car cant drive as well as a human" not "what the car needs to meet Lx", which are quite different things.
 
should Indian Govt standardize to better traffic rules?
I wanted to make a general comment about this.

It is not about what the rules are. When it comes to developing countries, it is mostly about enforcement. Whether it is about sweat shops or child labor or bad traffic. It is almost always never about the rules - because it is easy to make rules (well, we are not talking about US Congress !), but consistent enforcement is another matter.

And its not just about developing countries either - we have a ton of examples in US itself.
- Almost every car drives over the "max" speed limit. Why ?
- There is widespread use of rolling stops ...
- Don't even get me started on all kinds of illegal discriminations ...
 
Your "side camera" idea continues to be nonsensical, the fender cams look straight back, you can't angle the car that much at a turn that's obscured to the B-pillar to be useful without either already being in the flow of cross traffic or being so far angled you're blocking 2 lanes doing it.
DFh2Vv4XUAIVStb.jpg


Look at the left rear side camera. It's not "straight back" as you nonsensically claim. Turn it 30degrees clock wise and it will cover traffic from the left.
 
  • Like
Reactions: elasalle
DFh2Vv4XUAIVStb.jpg


Look at the left rear side camera. It's not "straight back" as you nonsensically claim. Turn it 30degrees clock wise and it will cover traffic from the left.

it was 45 degrees last time you told this story- now it's 30? Can you show your math?

As to what you see, I mean, it mostly IS straight back, that's the view you get when you activate the turn signal... but let's ignore that for a second...



Now show me an actual real world intersection where my car can be ALL of these at the same time:

Turned 30 degrees to the side right at the light/stop sign (but pointing straight while I am the second car in line so I can GET to the light/stop sign)
AND
I'm not puling into oncoming traffic doing so
AND
I'm not at least partly another lane with the front or rear of the car


Outside maybe the cybertruck, Teslas can't tank turn in place my dude.
 
it was 45 degrees last time you told this story- now it's 30? Can you show your math?
30 degrees is less than 45.

Here is the math:
30<45 = TRUE

My argument is:
1. The rear side camera is in front of the driver leaning forward
2. Turning the car 30degrees in an intersections is possible
3. When you turn the car 30degrees the rear side camera covers side traffic

1,2,3 -> rear side camera position is sufficient for right hand turns.

Which part of do you disagree with?
 
Last edited:
30 degrees is less than 45.

Here is the math:
30<45 = TRUE


Why are you showing math on things nobody asked about?

I asked to show the math that 45 (your original claim) vs 30 (your new claim) actually solves the problem.


My argument is:
1. The rear side camera is in front of the driver leaning forward

Sure. With a narrow FOV that mostly faces backward-- we know this because that's what the turn signal cam shows (even your own graphic shows how narrow that view is)


2. Turning the car 30degrees in an intersections is possible

This is (largely) the part of the argument I just asked you to prove.

Here's a reminder since it was a whole *checks notes* 2 posts ago.


Now show me an actual real world intersection where my car can be ALL of these at the same time:

Turned 30 degrees to the side right at the light/stop sign (but pointing straight while I am the second car in line so I can GET to the light/stop sign)
(Outside maybe the cybertruck, Teslas can't tank turn in place my dude.)
AND
I'm not puling into oncoming traffic doing so
AND
I'm not at least partly another lane with the front or rear of the car
 
Why are we relying on drawings of camera views? We have video extracts from all camera views via Green from several years ago. You can see exactly when cars travelling toward the Tesla leave the forward facing view and enter the B-pillar view. In my opinion, the camera coverage is a full 360, and barring occlusions, is sufficient.


 
  • Informative
Reactions: jerry33
Why are you showing math on things nobody asked about?

I asked to show the math that 45 (your original claim) vs 30 (your new claim) actually solves the problem.
If 30 degrees fixes the camera angle then 45 degrees also does it... It's a harder requirement to do it with a smaller angle.

This is (largely) the part of the argument I just asked you to prove.
2. Turning the car 30degrees in an intersections is possible


qiPUbkx.png


I will argue that the car is ~30degrees turned. If slightly less then I argue that it could have hugged the left lane marking a lot more and turned at a much sharper angle if needed. I will guesstimate that the rear side camera is looking 90° to the side when the car is ~10° turned and I think another 10° is needed for the video to cover cars on a collision course with our future trajectory.
 
Last edited:
If 30 degrees fixes the camera angle then 45 degrees also does it... It's a harder requirement to do it with a smaller angle.

Right- I'm asking, since you're tossed out 2 different angle #s so far, how do you know either is correct? What math did you use to conclude how much angle fixes the issue?



This is (largely) the part of the argument I just asked you to prove.
2. Turning the car 30degrees in an intersections is possible

You then follow up with a picture of 2 vehicles, it's unclear which you're talking about, but which includes no markings of the angle of either-- and the one that hasn't already entered the intersection appears to be in some kind of lane nearly double the width of some typical local roads lanes (though again with no labeling or scale it's hard to judge.



I will argue that the car is ~30degrees turned.

Can you try proving your claim instead of "arguing" it?


I will guesstimate that the rear side camera is looking 90° to the side when the car is ~10° turned and I think another 10° is needed for the video to cover cars on a collision course with our future trajectory.

So now you're down to 20 degrees angle? (10 plus another 10) down from 30, which was itself down from 45) all without showing your math how you "guesstimate" any of it?



Why are we relying on drawings of camera views? We have video extracts from all camera views via Green from several years ago. You can see exactly when cars travelling toward the Tesla leave the forward facing view and enter the B-pillar view. In my opinion, the camera coverage is a full 360, and barring occlusions, is sufficient.

But that's camera views on an unobstructed open road. Which isn't the issue at all.

Go back and look at the Chuck Cook video to see the obscured intersection situations we're actually discussing.

The "360 view" only shows you trees, bushes, or a fence instead of oncoming parallel traffic using the only camera facing that direction because it's stitting all the way back at the B-pillar.
 
So now you're down to 20 degrees angle? (10 plus another 10) down from 30, which was itself down from 45) all without showing your math how you "guesstimate" any of it?
In the first example(chuck driving through intersection) I said it was possible to aim the car 45°. Then I say 20° is enough(based on the sensor array). Then I gave an image example of a ~30° with lots of room to spare from the manoeuvre. It should not be impossible to imagine that the car could have turned sharper if that was the issue.

You still want to argue that there was no possibility to turn right with the rear side camera covering the traffic from the left in the situation with chuck?

I will quote you from the main thread.
Elon himself admits he's sometimes wrong--- why can't you?