Autonomous Car Progress

[ewoodrick]

You did.

If the car had crashed, you’d be legally responsible. If you had gone too fast, you’d be fined.

Same thing as when a FSD or AP car crashes and Tesla says « Tesla is not responsible, it is very clear per the license agreement that the driver must remain attentive at all time and ready to take control ». You cannot have it both ways.

Doing something and being responsible for it are two very different things.

If I'm learning to fly a plane and don't have my license and I'm with a Flight Instructor and we have an accident while I am flying and he is not flying the plane, he's responsible. But he wasn't flying.

 

[ewoodrick]

I literally quoted Tesla telling you that

Here it is again.

Tesla said:
we do not expect significant enhancements in OEDR or other changes to the feature that would shift the responsibility for the entire DDT to the system. As such, a final release of City Streets will continue to be an SAE Level 2, advanced driver-assistance feature.

Do you think Tesla is lying?

Is City Streets the EXACT SAME THING as FSD Beta or FSD? I don't think so.

Not lying Apples do not equal oranges.

 

[EVNow]

Come on, if you are going to throw out some facts, give a little more details. How many are participating in your crowdsourced statistics? Are these user reported or automated?

It wouldn't happen to be the 300 person project would it?

Come on - I've posted the links multiple times. Look it up.

Its better than nothing - because thats all you got from Tesla. I've also posted my personal stats. Have you posted your stats - do you record all your drives ... ?

 

[Knightshade]

Is City Streets the EXACT SAME THING as FSD Beta or FSD? I don't think so.

Yes, but as we've seen throughout the thread- what you think is factually wrong and no amount of direct quotes from CA DMV, NHTSA, and Tesla themselves telling you you're wrong seems able to make you reconsider.

Which has gotten increasingly hilarious

I'm sorry, I was thinking

In court a lawyer might call this assuming facts not in evidence.

I didn't think that Elon had telekinesis working yet.

I know that I wasn't turning the wheel.

Elon doesn't need to turn the wheel. Neither even do you.

As you've had repeatedly pointed out to you, lateral control (steering) tells you NOTHING about the SAE level of a car other than "not level 0"

Every level from 1 and up could have lateral steering control--- and not until you get to L3-- which has a number of OTHER requirements including ones Teslas system lacks- is anyone but the human behind the wheel driving the car.

 

[hgmichna]

Doing something and being responsible for it are two very different things.

If I'm learning to fly a plane and don't have my license and I'm with a Flight Instructor and we have an accident while I am flying and he is not flying the plane, he's responsible. But he wasn't flying.

"Flying" a plane, or "driving" a car, is not well-defined. Avoid such expressions or you will have endless arguments between people assuming a popular or a legalistic definition. We've had that here time and again.

If the Tesla Autopilot is active, is it driving the car? Many will say yes, but quite a few will vehemently insist on no, because of the legalistic definition. The result is a senseless argument.

Avoiding it is actually quite simple. Avoid the word drive and simply say the autopilot is turning the steering wheel, accelerating, and braking, with a supervising human ready to step in. And the whole argument collapses, because it is not a factual one, but only one about the ambivalent meaning of a word.

I had said this here before and will repeat it occasionally.

 

[ewoodrick]

"Flying" a plane, or "driving" a car, is not well-defined. Avoid such expressions or you will have endless arguments between people assuming a popular or a legalistic definition. We've had that here time and again.

If the Tesla Autopilot is active, is it driving the car? Many will say yes, but quite a few will vehemently insist on no, because of the legalistic definition. The result is a senseless argument.

Avoiding it is actually quite simple. Avoid the word drive and simply say the autopilot is turning the steering wheel, accelerating, and braking, with a supervising human ready to step in. And the whole argument collapses, because it is not a factual one, but only one about the ambivalent meaning of a word.

I had said this here before and will repeat it occasionally.

I disagree. Flying a plane is extremely well defined and litigated. And the use of autopilot in a plane is quite defined as well. (Although a plane autopilot has an infinitely easier job)

Flying has been as extremely litigious industry over the years, many manufacturers had to close up because of it. But legislative relief has occurred, and new things are starting to be introduced again.

 

[Knightshade]

I disagree. Flying a plane is extremely well defined and litigated.

So is driving (Well, extremely well defined anyway-- quite a few US states for example all have laws specifically addressing automated driving systems and what they need to be legally considered to be driving the car-- with all of them leveraging the language of SAE J3016-- there hasn't been much litigation over it yet though)

And Teslas system can't do it. It is, legally and functionally, an ADAS system that is incapable of meeting the legal requirements to actually drive the car

As Tesla themselves point out in the CA DMV emails- even specifically citing some of the things their system lacks to be capable of doing so. Where they also point out the fact that fsdb (which is also called city streets- both by Tesla, CA DMV, and the NHTSA since they all understand they're the same thing) is not intended to EVER gain that functionality, and anything >L2 will be a future development project.

 

[ewoodrick]

So is driving (Well, extremely well defined anyway-- quite a few US states for example all have laws specifically addressing automated driving systems and what they need to be legally considered to be driving the car-- with all of them leveraging the language of SAE J3016-- there hasn't been much litigation over it yet though)

And Teslas system can't do it. It is, legally and functionally, an ADAS system that is incapable of meeting the legal requirements to actually drive the car

As Tesla themselves point out in the CA DMV emails- even specifically citing some of the things their system lacks to be capable of doing so. Where they also point out the fact that fsdb (which is also called city streets- both by Tesla, CA DMV, and the NHTSA since they all understand they're the same thing) is not intended to EVER gain that functionality, and anything >L2 will be a future development project.

I'm tired.

Go sue Tesla for well, whatever they are doing wrong. and leave us alone. You are dead set to believe that Tesla can never do L3 or above. We understand what YOU believe. And I know what I know.

We all agree, you don't believe that Tesla will ever do L3 or above, We all below that you believe that you have the official documentation and interpretation of that documentation to assure that Tesla will never do L3 or above.

We all are wrong, and you are indeed correct. All of us that bought FSD have clearly wasted all our money. and well we are all screwed.

Tesla will never do L3, L4, nor L5 as per SAE automation guidance.

 

[Knightshade]

I'm tired.

Yeah, being wrong this much must be exhausting for you

Go sue Tesla for well, whatever they are doing wrong. and leave us alone.

...what?

On the contrary, I'm the one telling you what Tesla is actually doing and saying and you keep telling us they must be lying and that what Tesla is saying must be wrong.

You are dead set to believe that Tesla can never do L3 or above.

This is an outright lie, and I've never said any such thing at all.

What I said was fsdb can never do L3 or above.

Because that is literally what Tesla told a government agency when they asked

They went on to add (and I made sure to mention- numerous times) they intended to develop a FUTURE thing that DOES do L3 and above.


So once again you're insisting I said something that is the exact opposite of what I actually said


For example, this is what I posted back on Thursday

And Tesla has told us, in writing to a government agency, their software not only does not have those capabilities, they have no plans whatsoever to include them in the final version of FSDb.

They go on to say they DO plan to develop them in a FUTURE product. That is not FSDb.

Because FSDb is, and is always intended to remain L2.

I repeated this several times in the days since-- and yet here you are insisting I said exactly the opposite of what I actually said.

it's remarkable.

We understand what YOU believe.

Clearly you do not.

And I know what I know.

That's gotta be a short list.

We all agree, you don't believe that Tesla will ever do L3 or above,

Again- that is the exact opposite of what I actually said

And this latest post of yours is no different.

 

[ewoodrick]

What I said was fsdb can never do L3 or above.

Just so I am following you correctly.

1) FSD Beta will be expanded to support L3 or above?
2) FSD Beta will need to be completely rewritten to support L3 or above? Everything today thrown out.

 

[Knightshade]

Just so I am following you correctly.

You have failed to do so throughout the entire discussion. No exception to be found in this post either.

1) FSD Beta will be expanded to support L3 or above?

No. Once again you're making up nonsense I never wrote.

fsdb, which is the same thing as city streets- as confirmed by direct words from Tesla, CA DMV, and the NHTSA, is an L2 product. it was designed and intended as one, and the final version of it will remain L2 and incapable of higher operation. PER TESLA.

2) FSD Beta will need to be completely rewritten to support L3 or above? Everything today thrown out.

Again making up complete nonsense I never wrote.

Instead I wrote that there'd be some future developed software that would be designed with intent to operate at L3 or higher, but that software is not fsdb.

The reason I wrote that is because Tesla literally said that


I also mentioned that it would likely leverage much of what they learned and developed with fsdb-- but of course would require quite a bit more in addition since, as Tesla tells us, there's quite a bit lacking in fsdb for achieve >L2. Which only makes sense because the parts of the DDT that fsdb can do well, even if it can't do other parts, would make perfect sense to flow into the new product. There'd be no reason to "throw out" those parts- it's just another thing you invented in your head that nobody claimed.



I think the problem is- rather than try and read and understand what people (including Tesla) are actually saying-- you've made up your own entire story in your head.

And you're really bad at making up stories.

So you keep getting confused when they don't make sense.

 

[Bitdepth]

And you're really bad at making up stories.

So you keep getting confused when they don't make sense.

I think the issue for some people is that they believe the feature they have now FSD beta is an update away from actual L3 - L5 ADS. You can attribute that on the CEO and Tesla selling that narrative.

 

[Knightshade]

I think the issue for some people is that they believe the feature they have now FSD beta is an update away from actual L3 - L5 ADS. You can attribute that on the CEO and Tesla selling that narrative.

Almost certainly, yes.... Elon has been far more big picture and...overly optimistic... but every time we get to see anything internal from the people actually working on this the story is quite different.

One of the great quotes from Tesla about this was in those DMV emails

Elon's tweet does not match engineering reality

 

[ewoodrick]

You have failed to do so throughout the entire discussion. No exception to be found in this post either.


No. Once again you're making up nonsense I never wrote.

fsdb, which is the same thing as city streets- as confirmed by direct words from Tesla, CA DMV, and the NHTSA, is an L2 product. it was designed and intended as one, and the final version of it will remain L2 and incapable of higher operation. PER TESLA.


Again making up complete nonsense I never wrote.

Instead I wrote that there'd be some future developed software that would be designed with intent to operate at L3 or higher, but that software is not fsdb.

Sorry my intent was for you to confirm if either question is true from your standpoint.

So I'll try it a different way.

What is Tesla going to have to do to get beyond L2?

I really am lost, because when I asked if they will be able to expand the existing FSD Beta code, it sounded like you said no.
So I thought that the other option would be that they would have to rewrite all the code, for which you seemed to say no to as well.


What is Tesla going to have to do? I really want to hear your input. A few lines with no quotes of SAE. What steps will Tesla be taking to create code to go pass L2?

 

[Knightshade]

Sorry my intent was for you to confirm if either question is true from your standpoint.

So I'll try it a different way.

What is Tesla going to have to do to get beyond L2?

Write actual self driving software, instead of an ADAS system (which is what fsdb is)

I really am lost, because when I asked if they will be able to expand the existing FSD Beta code, it sounded like you said no.

Probably because Tesla, also, said no. Explicitly.

They plan to develop something, that is not fsdb, in the future, that would, offer L3 or greater capabilities.

So I thought that the other option would be that they would have to rewrite all the code, for which you seemed to say no to as well.

Why in the world would they need to do that?

Recycling parts of your code when doing a new project has been a tradition in programming for...basically ever.

Video game companies, for example, often take the code from a previous game- sometimes just parts- sometimes the entire game engine- and then use it as the basis for a future-- more advanced-- game to which they then add additional things in said new product.

It's pretty rare anybody throws "everything" out

The fact, to pick one specific example, the vision code recognizes and reads speed limit signs does not grant any particular level of driving automation- but it's useful, working, code... so I'd certainly expect them to use it in the new >L2 software... though they'll obviously need to build further so it can handle more complex signs it currently does not. Tons of examples of things like that in the current code that'd be moved at least in part to a new system.

What is Tesla going to have to do? I really want to hear your input.

For level 3? Write software with a complete OEDR.

Which fsdb, even in its final version, is not.

They'd also likely need to improve significantly some of the current things that ARE in fsdb.

For L4 they'd need that PLUS robust DDT fallback code to eliminate the need for a human at all in the system.

This all assumes the hardware is capable of the job of course... HW3 almost certainly is not. It's unknown if HW4 will be-- there's certainly problematic blind spots in the HW3 system...problematic for parking, and for some poor-visibility intersections as well... and it's unclear if those are "fixed" with 4 or not.


Since nobody in the world has developed a working L5 self driving system (or even a general, not massively restricted and geofenced, L4 one) nobody actually KNOWS how much hardware you need to achieve it, regardless of your software.

 

[ewoodrick]

So I guess what you are saying is that they can't use the FSD beta code, but they can use the parts and pieces of the FSD Beta code. But they probably can't use HW3. Is that right?

 

[Knightshade]

So I guess what you are saying is that they can't use the FSD beta code, but they can use the parts and pieces of the FSD Beta code.

Bethesda software forked the Fallout 3 code (which was built on the Gamebryo engine) to make the Creation engine they used for Skyrim (and then modded further for Fallout 4 and Fallout 76).

None of those last 3 were Gambryo, but they reused a decent amount of code FROM it.

No reason whatever the >L2 software Tesla develops won't do that with fsdb code....borrowing some amount of it to develop a new thing.

But they probably can't use HW3. Is that right?

Probably. We know they had already run out of spare compute back in 2020-- at that point what was originally intended to be 2 redundant nodes for self driving was turned into 2 nodes with a single instance of code spanning both and no redundancy at all.... and quickly expanded to fill up all that compute as well.

As I suggest, nobody actually knows how much compute is "enough" for legit vision-based self driving- nobody including Tesla- and nobody will known until they actually achieve it... so HW4 might be enough- or it might not.... but it's a pretty good bet HW3 isn't going to get there.

That's apart from the limitations of the HW2-3 era cameras both in resolution and placement/field of view.... possibly we will gain more insight into Teslas thoughts on that when we get a better/deeper analysis of the HW4 camera suite that appears to be starting to be delivered to customers now on some S/X vehicles maybe-- but also is being delivered with NO FSD features at all at the moment.

 

[jabloomf1230]

Not only does Elon's view of autonomous driving not match engineering reality, no other AD company's view does either. Sure, Waymo, Cruise, Mobileye etc. can run expensive sensor arrayed cars within some small geofences, but there's no evidence that this approach will scale to an economically viable service for hundreds of large metropolitan areas worldwide.

 

[EVNow]

Not only does Elon's view of autonomous driving not match engineering reality, no other AD company's view does either. Sure, Waymo, Cruise, Mobileye etc. can run expensive sensor arrayed cars within some small geofences, but there's no evidence that this approach will scale to an economically viable service for hundreds of large metropolitan areas worldwide.

Infact its almost guaranteed most of the AV companies will end up losing a huge amount of money. Even Waymo may never turn a profit.

 

[jabloomf1230]

Infact its almost guaranteed most of the AV companies will end up losing a huge amount of money. Even Waymo may never turn a profit.

I wouldn't say never. But reality is setting in for proponents of both Elon's everyman robotaxi and the rest of the field's geofenced rodeo.

Basically, the hard part for any autonomous car software is predicting what the next nearby nutty human driver will do. It got me to thinking. At some point it might be helpful to add an option to Tesla's UI to color code the "blue cars" based on the NN's view of risk. A range of colors might help the driver in gaining a better understanding how FSD works.