Autonomous Car Progress

[banned-66611]

No, they didn't.


See attached for that they actually promised. It's easily read as L4.



The website didn't.

Again "what musk said" and what "the company told you during purchase" are different things legally.




WITH a driver in the car. And charging itself has nothing to do with driving levels either.

Notice where it says "almost all circumstances"

That's the disclaimer to get you L4, not 5.

Notice where it mentions person in driver's seat, and person stepping out of car?

View attachment 559555

It says clearly that it will go off and park itself with no driver. It doesn't mention the person in the driving seat needing a licence either. Does not mention any geographic limits.

That's L5.

 

[banned-66611]

Car drives you "almost anywhere" with nothing required of human in seat is 100% L4 at a minimum though.

It'd be 5 if they took out "almost"

In any case they are years and years away from L4 so it's a moot question. Robotaxi isn't launching this year, another year goes by without delivering what customers already paid for.

 

[Knightshade]

It says clearly that it will go off and park itself with no driver. It doesn't mention the person in the driving seat needing a licence either. Does not mention any geographic limits.

That's L5.

No, it's not.

L4 vs L5 doesn't care about the driver being there- neither "requires" a human to ever do anything when they're operating.

Waymo has L4 cars with NO driver at all running today.

And yes it mentions limits- what do you think "ALMOST anywhere" means?

L5 goes everywhere

L4 goes some set of places that is less than everywhere.

 

[banned-66611]

Elon's ability to estimate time to market has improved by at least an order of magnitude over the past couple years.
Plus, everything I am seeing from Karpathy et al is pretty much "we are on the right path and we see progress".

If they were thinking "oh crap, we chose the wrong technology, Waymo is years ahead and we are going to miss the robotaxi launch this year" do you think they would tell you?

 

[mikes_fsd]

You are more optimistic that me. Being "on the right path and seeing progress" does not guarantee that a product will be released by a certain deadline.

I am more optimistic than you, but that's not a big stretch.
I've also seen my first Tesla AP1 suck at taking flyover bridges that curved and within months seeing it get improved with OTA's.
Then with Model 3 FSD struggle to reach parity and then seeing NoA delivered and recently the traffic light detection and reaction all with OTA updates.

So, my optimism seems justified to me.
Can you imaging Waymo telling the world that they are rewriting the entire stack of their self driving cars?? No? neither can I, Elon on the other hand was told (I am guessing here) by Karpathy and crew that their software stack is whack and not scalable and presented a solution and Elon said go for it and then casually announced it in a podcast.

That one example is enough for me to be more optimistic about Tesla, not less. Because in my eyes they are not willing to go with a failed/flawed design just because a rewrite would slow them down initially by 6, 12, 18 months.

So, yes I am more optimistic than you, but I also have a 10 year daughter that I would love to be able to hand over the Model 3 with a pretty good feature set.

 

[DanCar]

In any case they are years and years away from L4 ...

Agree. I'd be surprised if Tesla delivers L3 beyond traffic jam situations on freeways in less than 2 years.

 

[daniel]

... Also, notice the caveat in Elon's tweet at the end where he says "it's a question of what functionality is proven to be safe".

And that's the rub. It's pretty much meaningless to have "features" in place if they are not safe enough to be used in consumers' cars.

 

[Tikib2020]

I think some of the people here having been burned by Elon before are too pessimistic about the timeframe.

I can see Tesla having feature complete FSD by the end of this year and robotaxis by the end of next year.

 

[diplomat33]

I think some of the people here having been burned by Elon before are too pessimistic about the timeframe.

I can see Tesla having feature complete FSD by the end of this year and robotaxis by the end of next year.

A robotaxi is more than just stopping at traffic lights and making turns at intersections. A robotaxi has to be able to handle every single driving case in its area with no human in the car. And a robotaxi has to be able to handle any and all unexpected situations with no human input. It just seems unlikely that Tesla can achieve that in just 16 months from now, especially considering that the rewrite is at least 2-4 months away according to Elon.

 

[daniel]

I think some of the people here having been burned by Elon before are too pessimistic about the timeframe.

I can see Tesla having feature complete FSD by the end of this year and robotaxis by the end of next year.

"Feature complete" is such an arbitrary term that I'll make no predictions or place any bets. They need to get NoA working in the city at a beta level, but might declare it long before it's reliable, just as NoA on the highway was reportedly quite unreliable when they first released it, and many people still find it to be unreliable.

As diplomat33 points out above, getting from a Level 2 beta feature to one that is mature enough to operate driverless is a huge step. Note that all the features of EAP are still described by Tesla as beta features. All these years later, still beta. And all these years later, they still require some driver intervention and still have some failures. (The most common one for me is on a stretch of road that goes from two lanes to one in my direction and I need to merge left. Sometimes auto-merge works and sometimes it doesn't. Not a big deal, but would be unacceptable for FSD. The lanes are clearly marked, but sometimes it sees them and sometimes it doesn't, for no obvious reason.)

Tesla is at least 5 years away from robotaxi anywhere but possibly some extremely narrowly defined geographic area, and I don't think they're ever going to get to non-geofenced robotaxi at all with the current sensor set.

 

[Knightshade]

A robotaxi is more than just stopping at traffic lights and making turns at intersections. A robotaxi has to be able to handle every single driving case in its area with no human in the car. And a robotaxi has to be able to handle any and all unexpected situations with no human input.

I mean technically Waymo is operating robotaxis that don't do that, so it's not a requirement.

Though I don't see Tesla setting up a remote driver assist center like Waymo uses.

 

[diplomat33]

I mean technically Waymo is operating robotaxis that don't do that, so it's not a requirement.

Though I don't see Tesla setting up a remote driver assist center like Waymo uses.

Certainly, Waymo does not yet have robotaxis that can handle every single driving case everywhere. But Waymo does have robotaxis that can handle all the driving in a geofenced area with no human input for quite awhile before they encounter a problem.

 

[DanCar]

But Waymo does have robotaxis that can handle all the driving in a geofenced area with no human input for quite awhile before they encounter a problem.

Sounds exaggerated to me. Here is how I would state Waymo:

1. Does lots of incredible things. Can drive around most of the time without driver input. Safety driver used 99.9% of the time.
2. Waymo has produced a few video's driving unassisted on a predefined route in broad daylight with a passenger, without an employee on board.
3. It has been 6 months since the last waymo driverless video.

 

[willow_hiller]

Sounds exaggerated to me. Here is how I would state Waymo:

1. Does lots of incredible things. Can drive around most of the time without driver input. Safety driver used 99.9% of the time.
2. Waymo has produced a few video's driving unassisted on a predefined route in broad daylight with a passenger, without an employee on board.
3. It has been 6 months since the last waymo driverless video.

I know Waymo has some sort of pilot ridehailing service (Waymo One), but there does seem to be an odd lack of videos or even written accounts of people who have used it.

Does anyone know if those authorized to use the app have had to sign NDAs?

EDIT: Found the answer myself, there is indeed an NDA required to use the app Waymo robotaxi app arrives on the App Store – TechCrunch

 

[diplomat33]

I know Waymo has some sort of pilot ridehailing service (Waymo One), but there does seem to be an odd lack of videos or even written accounts of people who have used it.

Does anyone know if those authorized to use the app have had to sign NDAs?

EDIT: Found the answer myself, there is indeed an NDA required to use the app Waymo robotaxi app arrives on the App Store – TechCrunch

Only the early access program requires a NDA. "Waymo One" which is the ride-hailing service for the general public does not require a NDA.

From your own article:

"Riders will first be invited to join the Waymo early rider program, which requires signing a non-disclosure agreement, before moving into the public arm of Waymo One. Once on Waymo One, participants can invite guests and talk publicly about their experience."

EDIT: I would also point that there are plenty of videos of Waymo rides pre-covid. Obviously, with covid, things temporarily shut down. Also, since Waymo has L4, the progress now is more incremental, improving reliability, not adding new features. So I don't think any videos from say last month would be drastically different from a video 6 months ago. The key will be seeing videos or getting reports of Waymo expanding to new areas. Waymo is deploying Jaguar I-Pace robotaxis with their latest generation of FSD hardware. Getting more news on that will be more useful IMO.

 

[diplomat33]

It has been 6 months since the last waymo driverless video.

And it's been well over a year since the last Tesla FSD demo.

 

[aymelis]

Those simulations are limited to the creativity of the mind to come up with situations that cars might see.

Sorry, I disagree with that. Simulations can introduce randomness in such a way that they can cover edge cases. So you can give the simulator a whole bunch of sequences, either written manually or extracted from the real world, and assemble all these sequences in random order, random timings, etc... And you can do that as many times as you want. For example you can mix an "infinite" number of curves, shadows, bicycles or pedestrians, occlusion of stop signs, other's or self speed, you name it...

 

[ZeApelido]

Sorry, I disagree with that. Simulations can introduce randomness in such a way that they can cover edge cases. So you can give the simulator a whole bunch of sequences, either written manually or extracted from the real world, and assemble all these sequences in random order, random timings, etc... And you can do that as many times as you want. For example you can mix an "infinite" number of curves, shadows, bicycles or pedestrians, occlusion of stop signs, other's or self speed, you name it...

No

 

[DanCar]

Sorry, I disagree with that. Simulations can introduce randomness in such a way that they can cover edge cases. So you can give the simulator a whole bunch of sequences, either written manually or extracted from the real world, and assemble all these sequences in random order, random timings, etc... And you can do that as many times as you want. For example you can mix an "infinite" number of curves, shadows, bicycles or pedestrians, occlusion of stop signs, other's or self speed, you name it...

Will it cover a sunset painted on a side of truck, and the vision system depth estimation making a mistake and plowing through the truck? The point being made is there is an infinite number of adversarial cases that can exist and simulation isn't going to cover them all.

 

[diplomat33]

Shanghai is launching a L4 self-driving taxi service for the public: