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.
Hmmm, I'm not sure they are meaningless. They are certainly subject to interpretation, but what isnt?

They're meaningless for gauging progress. Just the fact that both lane assist + TACC *and* 99.99% FSD can be level 2 is stupid. The SAE levels do nothing but confuse people and have many caveats that make it useless during discussion. Every time you mention any level, you have to also discuss the entire ODD, essentially describing the whole system again anyway.
 
  • Helpful
Reactions: mikes_fsd
They're meaningless for gauging progress. Just the fact that both lane assist + TACC *and* 99.99% FSD can be level 2 is stupid.


Why?

The current system rolled out to everyone always requires human supervision and ultimate responsibility 100% of the time.

So does the FSD beta.

That's exactly the definition of level 2 on the SAE scale.


The SAE levels do nothing but confuse people

Only the ones who don't take the time to understand them it seems.

It's not, at all, complicated.


and have many caveats that make it useless during discussion. Every time you mention any level, you have to also discuss the entire ODD, essentially describing the whole system again anyway.

Again not really.

ODD is not really relevant to what level it is, other than the distinction between 4 and 5 is the lack of an OD.


If it always has a human responsible, and does only ONE of steering OR braking to help the HUMAN driver then it's L1.

If it always has a human responsible and does BOTH steering and braking to help the HUMAN driver, then it's L2.

Simple.

If it sometimes requires a human to be responsible, and sometimes does not, then it's L3. Again ODD doesn't matter to the level just as in 1-2. ODD will tell you when each is responsible, but nothing in the ODD would change the level of automation.

Simple.


if it never requires a human but has an ODD then it's L4. Period. The details of the ODD don't change the level.

Simple.

If it never requires a human and works everywhere it's L5.

Simple.


Discussion of ODD can be useful to understand the detailed capabilities of either an L3 or L4 system specifically, but those details aren't relevant to the actual automation level of the car.
 
If it always has a human responsible and does BOTH steering and braking to help the HUMAN driver, then it's L2.

Simple.

It's not that simple as I pointed out, for two main reasons:

1) sae levels do not define safety requirements. Tesla can simply release the FSD beta as level 5, but we all know it'd crash every 30-50 miles or so, so that'd be stupid.

2) 99.99999% fsd can be level 2, so it's meaningless to use the levels to gauge progress. Lots of Tesla haters like to point out that Tesla is still level 2 and is therefore way behind some other company, when that's clearly not true.
 
Forget polls, just do some simple maths.

How much will it cost Tesla when their full self driving car kills someone or cripples them? How much is it worth to them to roll it out early? Once the latter exceeds the former... Look out.
FSD beta will require constant driver supervision. Theoretically it won't cost Tesla anything if car kills someone. Call it partial self driving if that makes it easier to understand. It costs people nearly a trillion dollars each year in car accidents.
 
Last edited:
  • Like
Reactions: mikes_fsd
It's not that simple as I pointed out, for two main reasons:

1) sae levels do not define safety requirements.

That's exactly what it's this simple.

You keep insisting the SAE levels don't provide useful distinctions of ability or progress.

They do, in fact, offer very clear levels of progress in capability, exactly as I laid out.

There's an inherent assumption that when a manufacturer certifies the system is capable of level X autonomy they mean SAFELY capable of it because they assume car makers are not sociopaths.

YMMV I suppose on that.

Tesla can simply release the FSD beta as level 5, but we all know it'd crash every 30-50 miles or so, so that'd be stupid.

Why do you not have any examples to support your argument other than ones you admit are stupid?


2) 99.99999% fsd can be level 2

This doesn't even make sense as a sentence.

If it's never capable of operating without a human actively driving, then it can't be higher than L3.

That's literally part of the definition you keep insisting is somehow unclear.

There's no "percentage" ever involved.

, so it's meaningless to use the levels to gauge progress.

Except, it's not.

L3 means it can finally do at least something without requiring active human driving

That would be a useful gauge of progress. Its' not there yet.

L4 means it never requires a human to be actively driving to operate safely. It's totally self-sufficient in it'd OD, and can safely park itself if it's going to leave said OD. No human needed ever.

That would be a useful gauge of progress. It's not there yet either.

L5 means L4, but works everywhere, no ODs involved.

That would be a useful gauge of progress. It's not there yet.



Lots of Tesla haters like to point out that Tesla is still level 2 and is therefore way behind some other company, when that's clearly not true.


I mean- it is true though.

Waymo operates L4 robotaxis in Arizona. They are significantly more capable in terms of self-driving than Tesla in that one place.

The difference is scalability.

Waymo has working L4 RTs today. But it's very hard and expensive to scale them to other places with the way they do it.

Tesla has no higher than L2 today, but is using a design philosophy that will work vastly more places and scale much better if it works out.
 
L3 means it can finally do at least something without requiring active human driving

Yup, you can have L3 that does very little (to the point of being useless) or essentially FSD. When you talk about any L3 car, you need to discuss the entire ODD.

The reason why I think the SAE levels are stupid is because they could have done so much more to help us understand the capabilities or advancement of a system. When we talk about any sae level by itself, it doesn't help us understand the system at all for the reasons I mentioned.
 
F1B6E821-5508-4CDD-B745-35D50246F706.gif
a very capable driver assistance system.

Elon’s reaction on your request to build you a very capable driver assistance system
 
Yup, you can have L3 that does very little (to the point of being useless) or essentially FSD. When you talk about any L3 car, you need to discuss the entire ODD.

This is, again, factually wrong.

Do it always require a human? If yes, it's not level 3.

Does it never require a human? if yes, it's not level 3.

Does it sometimes but not always require a human? If yes, it is level 3.


You don't need to know anything at all detail-wise about the ODD to determine what level the vehicle is.

ODD just tells you info about its capabilities within that level that you again don't need to know ODD info to assign to it.


The reason why I think the SAE levels are stupid is because they could have done so much more to help us understand the capabilities

That is literally not the point of the SAE levels.

it's to classify the level of automation by major milestone differences.

"needs a human all the time" 0-2 (with 0-2 distinguished by how many features, 0, 1, or 2, of "autosteer" and "accelerator/brake" are semi-automated.)

"needs a human some of the time" 3.

"needs a human never" 4-5 (with the only difference being "fully drives everywhere" or "fully drives less than everywhere")


it's a simple system you keep trying to make more complicated and assign jobs it's never meant to do.

It's also a very useful system for showing progress in automation if you understand what the actual levels mean.




When we talk about any sae level by itself, it doesn't help us understand the system at all for the reasons I mentioned.


But that remains fundamentally untrue.

It's apparently not helping you understand some far more nuanced questions it was never intended to address, but it's very useful for exactly the distinctions, differences, and demonstration of progress in automated capabilities I describe above.
 
automation by major milestone differences.

That's my point. The SAE levels aren't useful for gauging progress. Without knowing the details, you can't say a level 4 system (for example if the system is geofenced to a single route) is more advanced than a level 2 system (99.999% reliable fsd that still requires driver attention because the *mfg says so to absolve responsibility*).
 
  • Disagree
Reactions: beachmiles
That's my point. The SAE levels aren't useful for gauging progress.

But your point is factually untrue.

L2 always requires a human driver 100% of the time.

L3 sometimes does not require a human driver- that's clear and obvious progress.

L4 never requires a human driver- that's even clearer and more obvious progress.

L5 is L4 that works everywhere instead of some specific set of circumstances, so again clear and obvious progress.

Without knowing the details, you can't say a level 4 system (for example if the system is geofenced to a single route) is more advanced than a level 2 system

Except this completely misunderstands, again, the entire point of the SAE standard

It's not a "who is more "advanced"" scale.

it's a who is more automated scale.

A system that can drive without a human is more automated than one that can't. Period full stop.

That's what the SAE system is for

You can argue if one is more USEFUL than another, or more SCALABLE than another. But that's not the point of the SAE standard.

This might help you.

https://www.itu.int/en/ITU-T/extcoo...Update-on-Global-Ground-Vehicle-Standards.pdf

It lists the NUMEROUS SAE docs for automation. The one discussed so much here is J3016- its job is taxonomy for classifying driving automation features. That's it

It's not a spec document and not intended to be.

Your dissatisfaction continues to be caused by your refusal to understand what the doc is actually for.


There's many OTHER SAE specs that dive into more details-

For example J3164 is focusing on creating a taxonomy to codify behaviors and manouvers specific to L3 through L5 driving.

Or J3131 defining an ADS reference architecture with functional modules to support interfaces for L3 through L5 including terminology and best practices.

Or J3018 providing guidelines for safe testing of L3 through L5 prototype systems.

And even more listed there.[/QUOTE]
 
it's a who is more automated scale.

A system that can drive without a human is more automated than one that can't. Period full stop.

That's what the SAE system is for

Yup, I agree with you on that. I'm not saying the SAE levels are useless in every way, just for gauging progress, which is what many people in this forum use them for.

What we can say is that Tesla is developing and testing a level 5 system (possibly the ONLY level 5 system in advanced development, link us to another).
 
What's the absolute minimum required to call a vehicle Level 5? Let's say a start-up wants to game the system and release "The first Level 5 capable vehicle in the US" What would they need to make?

- Capable of driving on any road in the US, in any environmental condition
- Never needs intervention from a human driver, is capable of a safe pull-over in an emergency

Is there much else? Could it, for example, travel at 5 MPH at all times on every road, and run into obstacles frequently? It might still be Level 5, just a really rubbish Level 5.
 
What's the absolute minimum required to call a vehicle Level 5? Let's say a start-up wants to game the system and release "The first Level 5 capable vehicle in the US" What would they need to make?

- Capable of driving on any road in the US, in any environmental condition
- Never needs intervention from a human driver, is capable of a safe pull-over in an emergency

Is there much else? Could it, for example, travel at 5 MPH at all times on every road, and run into obstacles frequently? It might still be Level 5, just a really rubbish Level 5.
That would not be legal in most (all?) states. Even states that require no regulatory approval still require autonomous vehicles to obey the law and driving at 5mph everywhere is illegal.
 
  • Like
Reactions: powertoold
That would not be legal in most (all?) states. Even states that require no regulatory approval still require autonomous vehicles to obey the law and driving at 5mph everywhere is illegal.

It's also not legal to crash into other motorists, and I imagine some autonomous vehicles have done that before.

SAE doesn't say anything about the vehicle needing to operate in a legal manner, does it?
 
Yup, I agree with you on that. I'm not saying the SAE levels are useless in every way, just for gauging progress, which is what many people in this forum use them for.

But wat you're saying continues to be demonstrably untrue.

SAE levels are extremely useful for gauging progress of automation of driving

Which is the exact thing they're specifically written to do.


There's a significant different in automation progress as you move up each level.

Tesla has not gotten beyond 2 yet.

Others, perhaps most notably Waymo, have.

Again that doesn't mean Waymo isn't behind in other measures (like potential scalability, or consumer availability)



Y
What we can say is that Tesla is developing and testing a level 5 system (possibly the ONLY level 5 system in advanced development, link us to another).

We can say it but it'd be factually wrong.

Tesla has never suggested their system would be L5 (works in ALL circumstances).

If you think they have, please quote where they've done so.

Even today the aspirational description on the old FSD page they haven't updated in forever makes it clear it's an L4 intent- not 5 - when it says it will be capable of full self driving in

Teslas own description of future FSD functionality said:
almost all circumstances.

That "almost all" tells us L4, not 5.



What's the absolute minimum required to call a vehicle Level 5? Let's say a start-up wants to game the system and release "The first Level 5 capable vehicle in the US" What would they need to make?

- Capable of driving on any road in the US, in any environmental condition
- Never needs intervention from a human driver, is capable of a safe pull-over in an emergency

Is there much else? Could it, for example, travel at 5 MPH at all times on every road, and run into obstacles frequently? It might still be Level 5, just a really rubbish Level 5.


Again, 4 vs 5 is simply.

Level 4:

Vehicle never requires a human to perform the dynamic driving task.

Vehicle can do the entire dynamic driving task under some set(s) of circumstances (which can be very wide or very narrow)

Vehicle can safely park itself if it's moving outside those set(s) of circumstances.


L5:

L4 but under all conditions, period full stop.


In neither case is there any criteria for how well it does compared to the average human- there's simply the assumption of the SAE that car makers are not sociopaths not corporately suicidal and won't commercially put known murder machines on the road that just constantly run people over and whatnot.




It's also not legal to crash into other motorists, and I imagine some autonomous vehicles have done that before.

SAE doesn't say anything about the vehicle needing to operate in a legal manner, does it?


No, but the actual law does.

Nobody's going to jail for violating SAE taxonomy guidelines- they will if they sell an L5 car that breaks laws.
 
  • Disagree
Reactions: powertoold
Except they're not. I already gave good reasons why.

No you didn't.

You gave reasons that either ignored the actual text or purpose of the SAE document, or contradicted it.

You were instead shown specific ways in which the levels explicitly show differences in progress

And your only response to those specific examples appears to be "nu uh"

That's not an argument, it's pouting.


E
The levels are more about the mfg'ers acceptance of responsibility rather than the capability of the system.

No, they're not even remotely that at all.

Again you appear to have not read or understood either the purpose or contents of the document.