Autonomous Car Progress

[stopcrazypp]

Ford just announced that they have formed a subsidiary called Latitude AI to develop "eyes off" autonomous driving for consumer cars:



Here is the full press release: https://media.ford.com/content/ford...e-ai-to-develop-future-automated-driving.html

First, I find it odd that they would shut down ArgoAI just to create a new company to basically do the same thing, develop autonomous driving. If Ford wanted to focus more on L4 for consumer cars, why not just tell Argo to shift their focus especially since Argo already had advanced L4? This seems like a typical dumb legacy automaker move: shut down the company (Argo) that you had invested billions in and that could have given you L4 highway like you want, to likely put billions to start a new company to recreate the same thing.

Second, from the press release, they want to focus on highway driving and stop and go traffic. So it seems like latitude AI will likely focus on L4 highway and L4 "traffic jam".

Probably for the same reason why Cruise is not leading the effort to develop GM's Supercruise or Ultracruise for consumer cars. The company culture and mission is just at odds with the traditional car maker. The idea behind most robotaxi companies is to reduce the amount of cars that have to be owned, while traditional car makers want people to buy as much cars as possible, ideally multiple cars per household (even if they didn't "need" them). Most robotaxi companies also don't really care how their cars look to the consumer, they are quite happy to slap a whole bunch of sensors on a box shaped car and call it a day, while traditional automakers are quite stubborn about styling (which is why many still have grills on their EVs or try to make them look as close to their ICE counterparts as possible).

I remember after GM bought Cruise, they tried to get them to work on a system for GM's consumer products and the CEO of Cruise resisted (and thus was fired). Even today, from what I have read, Cruise only collaborates with the team working on GM's consumer ADAS in a limited way, but they are not integrated together and very few if any of Cruise's tech is getting into the consumer vehicles.

 

[JHCCAZ]

Mobileye has been in this business a long time. I bought one of the very first Volvo XC60 cars in May 2009 (2010 model) that had advanced features for the time. A very good adaptive cruise system, camera-based blind spot warning and audio lane departure warning, and the new City Safety feature, a low-speed vehicle and pedestrian collision avoidance (AEB) system.

I'm fairly sure that this Volvo feature package was largely partner engineered with Mobileye. However, my impression is that Volvo along with many other partner-customers have tried to become more independent. They may still be selling some features under Mobileye license, but have been going their own way in the advanced self-driving area. I don't know all the reasons, but I assume that cost and possibly licensing or engineering disagreements or the main problems. Apparently, it wasn't only Tesla that had difficulties with the Mobileye relationship.

It certainly is curious that more automakers aren't trying to get a jump on advanced ADAS using Mobileye, if they are as far along as the various presentations claim. You'd think that the big boys would be willing to offer expensive Mobileye tech in the short term, even if they want their own by next decade. Possibly the contractual terms are too constraining and/or Mobileye is particularly aggressive with legal action if a partner tries to support their own separate IP development. I'm only speculating.

 

[Knightshade]

It CAN BE AS SIMPLE AS ONE BIT!!!

Perform as L2 / Perform as L5

It can't be, of course, because the software lacks multiple, fundamental, capabilities that are required to exist for anything above L2.

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.


We've known this- direct from Tesla, for several years how.

How folks keep getting it wrong is baffling.

 

[AlanSubie4Life]

It can't be, of course, because the software lacks multiple, fundamental, capabilities that are required to exist for anything above L2.

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.


We've known this- direct from Tesla, for several years how.

How folks keep getting it wrong is baffling.

When Elon says it I will believe it. And only then.

 

[diplomat33]

It certainly is curious that more automakers aren't trying to get a jump on advanced ADAS using Mobileye, if they are as far along as the various presentations claim. You'd think that the big boys would be willing to offer expensive Mobileye tech in the short term, even if they want their own by next decade. Possibly the contractual terms are too constraining and/or Mobileye is particularly aggressive with legal action if a partner tries to support their own separate IP development. I'm only speculating.

I find this claim that automakers are not interested in Mobileye to be odd. Mobileye is putting SuperVision on 9 vehicles across 6 brands. And I think Mobileye will be announcing a deal with a major premium european automaker and a North American automaker later this year. And Mobileye says all these brands want to continue from SuperVision to Chauffeur. So Mobileye seems to be getting good traction with automakers.

 

[EVNow]

I find this claim that automakers are not interested in Mobileye to be odd. Mobileye is putting SuperVision on 9 vehicles across 6 brands. And Mobileye says all these brands want to continue from SuperVision to Chauffeur. So Mobileye seems to be getting good traction with automakers.

Until OEMs are ready to announce this feature with models and model years ... we have nothing much to go by.

 

[diplomat33]

Until OEMs are ready to announce this feature with models and model years ... we have nothing much to go by.

Fair enough. We will see how it actually plays out.

 

[2101Guy]

Which is why it’s interesting in the investor thread, many wonder why TSLA went down vs up after investor day. Sure lots of great things were promised but Wall Street knows the “when” is always the wildcard with Tesla.

“It’s all a little hard to believe, and coming from Tesla and Musk, it’s hard to take any of it seriously.”

Musk's Focus on World Domination Is Getting in Tesla's Way

At Tesla's 2023 Investor Day, Musk outlined his 'Master Plan 3' to eliminate fossil fuels worldwide — when all anyone wanted was news about the next Tesla cars.

jalopnik.com

 

[ewoodrick]

It can't be, of course, because the software lacks multiple, fundamental, capabilities that are required to exist for anything above L2.

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.


We've known this- direct from Tesla, for several years how.

How folks keep getting it wrong is baffling.

Just like turning FSD Beta on and off today. Double click it is on. Lift it is off. One button click NoA is on, click again it is off.

Not sure what you are disagreeing with something that is done every day.

You seem to be pretty anal about Tesla signing a piece of paper and checking an L2 or L3 or L4 or L5 box. The rest of us know that it is a bogus form.
No, this is not L3 or L4 or L5 software.

Yes, FSD beta is and always be L2 software.

BUT the FSD software will contain enormous amounts of the FSD Beta software.

Is software that controls the motor speed currently a part of the FSD Beta software? Sure. Will it be part of the FSD software stack? Sure.

You really are arguing over semantics that honestly only exist in organizations that haven't moved into the 21st century.

 

[Knightshade]

Just like turning FSD Beta on and off today. Double click it is on. Lift it is off. One button click NoA is on, click again it is off.

Not sure what you are disagreeing with something that is done every day.

You're making less sense with each post.

There is nothing to "turn on" in this case.

There is no code "just a bit flip away" that makes it more than L2.

Tesla explicitly said this.

You seem to be pretty anal about Tesla signing a piece of paper and checking an L2 or L3 or L4 or L5 box.

Literally nothing I wrote says anything like that.

The software entirely lacks features required to exceed L2. And will contnue to do so in final form. Per Tesla themselves.

The LEGAL stuff is an entirely separate conversation (though that TOO would need to happen for them to actually deploy anything >L2)

No, this is not L3 or L4 or L5 software.

I mean-- thanks for finally agreeing with what I said in the first place- but it makes your previous arguments about it all the more confusing.

Yes, FSD beta is and always be L2 software.

Again- correct- and what I posted quite some time ago.

BUT the FSD software will contain enormous amounts of the FSD Beta software.

Man you're on a greatest hits of stuff I already said a while ago and then you appeared to be trying to argue about for some reason but now agree with....

Weird.

 

[stopcrazypp]

I find this claim that automakers are not interested in Mobileye to be odd. Mobileye is putting SuperVision on 9 vehicles across 6 brands. And I think Mobileye will be announcing a deal with a major premium european automaker and a North American automaker later this year. And Mobileye says all these brands want to continue from SuperVision to Chauffeur. So Mobileye seems to be getting good traction with automakers.

AFAIK all of the brands that have been announced are smaller brands or Chinese brands that will not be much of a player in the western markets. It seems a lot of major western brands that previously used Mobileye for ADAS have parted ways and are going to other solutions for more advance L2 or beyond. I guess we will see what major brand on the US would go with SuperVision, but so far I don't believe any of them are.

 

[ewoodrick]

You're making less sense with each post.

There is nothing to "turn on" in this case.

There is no code "just a bit flip away" that makes it more than L2.

Tesla explicitly said this.

Literally nothing I wrote says anything like that.

The software entirely lacks features required to exceed L2. And will contnue to do so in final form. Per Tesla themselves.



Weird.

Wow, you completely miss my first point (because you don't seem to think that it is possible)
L2 software is L1 plus some stuff. L3 software is L2 plus some stuff. L3 is L2+ and so on. That's a simple fact of software. There's no SAE in that statement, just simple engineering. (well really complex software though)

Taken from the SAE site L2 to is L1 with both braking and adaptive cruise. L1 is "OR"
The difference between L3, L4, and L5 is really just reliability and "all conditions" (which many humans can't drive in "all conditions")

So the "turn on" that I am referring to is basically the enabling or disabling of specific features. I can turn on Adaptive Cruise Control and the car will be functioning in L1 mode. I turn it off and enable lane centering. It's still in L1. I turn both on then the car is functioning in L2. Pretty simple concept right from the SAE site.

You say that today's software lacks the features needed to "make it more than L2". If L2 is lane centering and adaptive cruise control, Then what is the ability to make turns?

So it really sounds to me like Teslas are currently running as L3 cars, as per SAE specifications above. Or do you not believe that a Tesla can "drive the vehicle under limited conditions and will not operate unless all required conditions are met?"

 

[ewoodrick]

ewoodrick said:
BUT the FSD software will contain enormous amounts of the FSD Beta software.

Man you're on a greatest hits of stuff I already said a while ago and then you appeared to be trying to argue about for some reason but now agree with....

Weird.

Pretty sure that you said that FSD Beta is not and will never be FSD.

 

[ewoodrick]

AFAIK all of the brands that have been announced are smaller brands or Chinese brands that will not be much of a player in the western markets. It seems a lot of major western brands that previously used Mobileye for ADAS have parted ways and are going to other solutions for more advance L2 or beyond. I guess we will see what major brand on the US would go with SuperVision, but so far I don't believe any of them are.

Yea it feels to me like more people have jumped off their bandwagon than jumped on it. It also seems clear that they really don't have an answer yet as well.

 

[Knightshade]

Pretty sure that you said that FSD Beta is not and will never be FSD.

And so did you.

Yes, FSD beta is and always be L2 software.

So I'm confused why you keep finding ways to argue with me when you appear to agree with what I'm saying.

Wow, you completely miss my first point (because you don't seem to think that it is possible)
L2 software is L1 plus some stuff. L3 software is L2 plus some stuff. L3 is L2+ and so on. That's a simple fact of software.

Sure.

But FSDb does not contain the "stuff" to go beyond L3.

There's no "switch" to throw and magically have that change. You can't just "turn on" the added capabilities because the added capabilities do not exist in the software. And the fsdb final product is not even intended to add them. Per Teslas own statements.

Some FUTURE product, that is not fsdb, will, Tesla hopes, be >L2. And as I've said for a long while now is likely to leverage a lot of what fsdb contains PLUS a bunch of new stuff it does not.

So the "turn on" that I am referring to is basically the enabling or disabling of specific features.

But the features needed for >L2 do not exist in the software

There's nothing to "turn on" with a flip of a bit.

Is telling you this a 5th time going to make you understand it? Likely not but hope springs eternal.

I can turn on Adaptive Cruise Control and the car will be functioning in L1 mode. I turn it off and enable lane centering. It's still in L1. I turn both on then the car is functioning in L2. Pretty simple concept right from the SAE site.

You say that today's software lacks the features needed to "make it more than L2". If L2 is lane centering and adaptive cruise control, Then what is the ability to make turns?

it's exactly the same as lane centering. Control of steering is a single thing, no matter what type of steering. It's lateral control of the vehicle.

I'd highly suggest you read the actual SAE spec instead of just the chart, because it explains all this in there.

So it really sounds to me like Teslas are currently running as L3 cars, as per SAE specifications above.

Dis you?

Yes, FSD beta is and always be L2 software.

Pick a lane my dude.


Anyway- FSDb is not level 3.

Per SAE definitions.

Tesla themselves explain the things the software is entirely missing to enable operation greater than L2 in the CA DMV emails.

Maybe go read those before you try posting further on this since you seem painfully unaware of Teslas own statements on this topic and how they directly tell you you're wrong.... (well...wrong the times you suggest it's >L2.... I suppose we have to give you credit for the times you contradict yourself and agree it's L3 and will remain so)

 

[diplomat33]

AFAIK all of the brands that have been announced are smaller brands or Chinese brands that will not be much of a player in the western markets. It seems a lot of major western brands that previously used Mobileye for ADAS have parted ways and are going to other solutions for more advance L2 or beyond. I guess we will see what major brand on the US would go with SuperVision, but so far I don't believe any of them are.

True. But Mobileye did say that they have a solid deal with a major premium European automaker and a major brand in the US. That should be more substantial.

 

[heapmaster]

Probably for the same reason why Cruise is not leading the effort to develop GM's Supercruise or Ultracruise for consumer cars. The company culture and mission is just at odds with the traditional car maker. The idea behind most robotaxi companies is to reduce the amount of cars that have to be owned, while traditional car makers want people to buy as much cars as possible, ideally multiple cars per household (even if they didn't "need" them). Most robotaxi companies also don't really care how their cars look to the consumer, they are quite happy to slap a whole bunch of sensors on a box shaped car and call it a day, while traditional automakers are quite stubborn about styling (which is why many still have grills on their EVs or try to make them look as close to their ICE counterparts as possible).

I remember after GM bought Cruise, they tried to get them to work on a system for GM's consumer products and the CEO of Cruise resisted (and thus was fired). Even today, from what I have read, Cruise only collaborates with the team working on GM's consumer ADAS in a limited way, but they are not integrated together and very few if any of Cruise's tech is getting into the consumer vehicles.

My guess is they did it for the $2.7 billion non-cash pretax impairment they recorded.

 

[ewoodrick]

But the features needed for >L2 do not exist in the software

The ability to drive on it's own is not in the software? Then what drove my car to lunch today, making the turns?

As it looks to me, L3 is anything above lane centering and Adaptive Cruise Control (L2).

Automatically turning at an exit sounds like L3 to me.

My 2018 (refreshed model) Leaf had lane centering and Adaptive Cruise Control. It wouldn't drive me to lunch.

 

[Knightshade]

The ability to drive on it's own is not in the software? Then what drove my car to lunch today, making the turns?

No, it does not have the ability to "drive on its own"

If you remain unclear why I again suggest you read Teslas own explanation of why in the CA DMV emails.

As it looks to me, L3 is anything above lane centering and Adaptive Cruise Control (L2).

Then you should probably read the actual J3016 document to understand the several ways in which your statement is factually wrong.

Automatically turning at an exit sounds like L3 to me.

it's not.

It's possible to have an L1 system that does that if the human controls speed. And obviously an L2 system can easily do it since that's what your car actually has. And of course L3, L4, and L5 can ALSO do that.


So "turning at an exit" tells you literally nothing about the SAE level of the vehicle other than "not level 0" since everything from l1 up could theoretically do that.


Again you need to go read J3016, not just the summary chart.

 

[ewoodrick]

No, it does not have the ability to "drive on its own"

Answer my question then. Who drove my car to lunch?