Elon: "Feature complete for full self driving this year"

[diplomat33]

Which could also be a convenient way of hiding the fact that they can't do this blended mode with humans in a timely fashion (especially w/o data volume). So it could be a punt in hopes that the other team (Tesla) fumbles before Waymo removes their steering wheel first? Or at least that could be their pitch to investors. "We're taking the safer route to full autonomy" I can imagine being said, which is clearly the endpoint in everyone's mind, hoping to leapfrog Tesla with this shortcut, and avoid lawsuits along the way. This makes the most sense in my head.

First, Waymo already had a L2 system, like Autopilot. You see in that video I posted that they had a AP system that worked really well on highways. So they absolutely could have deployed it in a timely manner if they had wanted to.

Second, there is a logical argument to be made that the blended human-machine driver assist is not a good idea. So why should Waymo still try to do something that is not a good idea? It makes more sense to focus on full autonomy which is better.

Also, Waymo is not an auto maker and they don't already have a fleet of L2 cars on the road. They don't have a product that they need to maintain or customers that they need to please. As a new company, they are free to just focus on a new product, ie full autonomy. Tesla, of course, does have a large fleet of L2 cars that they need to take care of. So it makes sense for Tesla to focus on adding L2 features to give their customers more value as they work towards full autonomy.

 

[J1mbo]

Waymo just moved the human monitor out of the driving seat and into a control room

 

[Daniel in SD]

Waymo just moved the human monitor out of the driving seat and into a control room

If they have the technology to do that they probably should dump all this autonomous crap and focus on that.

 

[Knightshade]

If they have the technology to do that they probably should dump all this autonomous crap and focus on that.

they can only do that because they also have all the autonomous crap on the vehicle as well.

the backup remote is human is for the narrower set of circumstances the car can't understand... (an example they gave was if the car sees a van being loaded up with side doors open blocking the only lane through- a human in the control room can judge "they look like they're leaving in a minute, just wait" or "they look like they will be here a while, turn around find another route"- while the car itself would have no way to "guess" how long a delay this is and if they should avoid it or not)

 

[RDoc]

So the human monitor camera is aimed at the backup driver in the control room to be sure (s)he's paying attention?

 

[Daniel in SD]

they can only do that because they also have all the autonomous crap on the vehicle as well.

the backup remote is human is for the narrower set of circumstances the car can't understand... (an example they gave was if the car sees a van being loaded up with side doors open blocking the only lane through- a human in the control room can judge "they look like they're leaving in a minute, just wait" or "they look like they will be here a while, turn around find another route"- while the car itself would have no way to "guess" how long a delay this is and if they should avoid it or not)

Exactly, the car has to recognize that there is an issue and ask the remote operator for help. You could have many more vehicles than remote operators depending on how often the vehicles get stuck.

So the human monitor camera is aimed at the backup driver in the control room to be sure (s)he's paying attention?

No.

 

[diplomat33]

Hope this is true. Could be a step closer to releasing some more AP features.

 

[diplomat33]

Hope this is true. Could be a step closer to releasing some more AP features.

Looks like it has been debunked by verygreen.

 

[banned-66611]

They are still struggling to get their image recognition to work.

 

[diplomat33]

They are still struggling to get their image recognition to work.

Yeah, based on what verygreen found, it appears that any "shadow mode" in the current build is just the FSD sneak preview visualizations. So right now, Tesla is working on the rewrite and the image recognition for traffic lights and other stuff. They need to get that right before they can do any "city driving".

 

[Snuffysasa]

Waymo just moved the human monitor out of the driving seat and into a control room

Not really true. The control room is NOT able to take control of the Waymo vehicle. They are unable to disengage and drive the vehicle, they are even unable to send an emergency brake command. And furthermore, it is not 1 human to 1 vehicle ratio of monitoring.

Hope this is true. Could be a step closer to releasing some more AP features.

Who is Third Row Podcast again? Do they have root access to a Tesla?

While I agree with verygreen, about the lack of shadow mode validating city streets driving. However, I'm sure they have been validating the performance of traffic light perception for quite some time now. Even like 9 months ago, Tesla said at share holders meeting that they are validating traffic light detection performance with the fleet vehicles.

Yeah, based on what verygreen found, it appears that any "shadow mode" in the current build is just the FSD sneak preview visualizations. So right now, Tesla is working on the rewrite and the image recognition for traffic lights and other stuff. They need to get that right before they can do any "city driving".

I am a little out of touch, what brought you to the conclusion that they are working on a rewrite of image recognition for traffic lights?

I was thinking there is a chance stopping at red lights / stop signs is the features that musk tweeted about recently. If musk says something is coming this month, then there must be some new feature that they are getting close to shipping.... (even if it is a few months away). I figured stopping at relights is a likely candidate. I know other people were speculating they wanted to release this feature before the end of march to count some FSD revenue in quarterly earnings or something like that.

 

[Knightshade]

I am a little out of touch, what brought you to the conclusion that they are working on a rewrite of image recognition for traffic lights?

Elon specifically said the entire AP code base was getting a fundamental re-write (which he claimed was "close" to being done)

I was thinking there is a chance stopping at red lights / stop signs is the features that musk tweeted about recently. If musk says something is coming this month, then there must be some new feature that they are getting close to shipping.... (even if it is a few months away).

He originally said enhanced summon was close to shipping November 2018.

It didn't ship until almost a full year later.


The FSD coast-to-coast drive he promised in 2016 to be ready by end of 2017 still isn't anywhere on the radar (pardon the pun)

 

[Snuffysasa]

Elon specifically said the entire AP code base was getting a fundamental re-write (which he claimed was "close" to being done)

right, I am wondering why we were concluding this was specifically about the image recognition of traffic lights.

He originally said enhanced summon was close to shipping November 2018.

It didn't ship until almost a full year later.

I realize that. And no doubt, we can't trust Elon timeframes. But in November 2018, they were certainly testing enhances summon and working on getting it released to the full fleet.

I was simply saying they must have a particular feature that they are currently trying to release to the fleet soon. Of course it could get delayed by several months.

 

[diplomat33]

Who is Third Row Podcast again? Do they have root access to a Tesla?

He's a Tesla investor and huge Tesla "fanboy" who has his own podcast where he promotes all things Tesla. He did an exclusive interview with Elon a couple months ago.

While I agree with verygreen, about the lack of shadow mode validating city streets driving. However, I'm sure they have been validating the performance of traffic light perception for quite some time now. Even like 9 months ago, Tesla said at share holders meeting that they are validating traffic light detection performance with the fleet vehicles.

I agree.

I am a little out of touch, what brought you to the conclusion that they are working on a rewrite of image recognition for traffic lights?

I don't think I said that they were rewriting traffic light recognition. I was just referencing the overall FSD rewrite that Elon mentioned awhile ago. And based on the FSD sneak preview, we know that they are running image recognition for traffic lights, stop signs, road markings etc... and validating their accuracy.

I was thinking there is a chance stopping at red lights / stop signs is the features that musk tweeted about recently. If musk says something is coming this month, then there must be some new feature that they are getting close to shipping.... (even if it is a few months away). I figured stopping at relights is a likely candidate. I know other people were speculating they wanted to release this feature before the end of march to count some FSD revenue in quarterly earnings or something like that.

Yes, I think there is a good chance we will see "stopping at red lights/stop signs" soon since traffic light and stop sign detection has been running for over a year now. So hopefully it is getting more reliable now. As soon as it is reliable enough, they will be able to push out the code for responding to the detection.

 

[Knightshade]

right, I am wondering why we were concluding this was specifically about the image recognition of traffic lights.

To me a fundamental re-write suggests it'll impact everything to varying degrees.

I realize that. And no doubt, we can't trust Elon timeframes. But in November 2018, they were certainly testing enhances summon and working on getting it released to the full fleet.

Sure. Just as we know they've been working on red lights for a while.

But it's possible they found out their approach wasn't reliable enough and this fundamental re-write will (they hope) fix that- and they won't know till it's done then test it.

I was simply saying they must have a particular feature that they are currently trying to release to the fleet soon. Of course it could get delayed by several months.

Sure. Or several years.

Neither would genuinely surprise me either way.

 

[Snuffysasa]

@Knightshade @diplomat33

Thanks, and agreed.

 

[EVNow]

He's a Tesla investor and huge Tesla "fanboy" who has his own podcast where he promotes all things Tesla. He did an exclusive interview with Elon a couple months ago.

Yes, 3rd row pod guys are what you would call "fanboys". The interview was just a wasted opportunity (in terms of FSD, though lot of useful history about Musks).

 

[banned-66611]

Elon specifically said the entire AP code base was getting a fundamental re-write (which he claimed was "close" to being done)

Which is why we can't believe a word he says.

When he said that robotaxi would be starting this year the re-write must have been started already, or at least on the cards. And a re-write means that not only are they back to square one on the FSD, they are back to square one on the current Autopilot features. They have to re-test and re-develop lane following, lane changing, the visualisation display, summon, emergency braking, vehicle following, everything.

And saying it's "close" to being done is simple misdirection. What is close to being done, version 0.1 Alpha of the re-write code that could never be released to the public? Or V1.0 where everything that the current AP does already works? And how long after that until FSD features can be added?

Tesla fans lap it up, expecting a pay-off on that big chunk of cash they handed to Tesla and a new toy to show off and impress their friends with. I just wish Elon would be honest about the state of AP and FSD.

 

[Battpower]

Which is why we can't believe a word he says.

When he said that robotaxi would be starting this year the re-write must have been started already, or at least on the cards. And a re-write means that not only are they back to square one on the FSD, they are back to square one on the current Autopilot features. They have to re-test and re-develop lane following, lane changing, the visualisation display, summon, emergency braking, vehicle following, everything.

And saying it's "close" to being done is simple misdirection. What is close to being done, version 0.1 Alpha of the re-write code that could never be released to the public? Or V1.0 where everything that the current AP does already works? And how long after that until FSD features can be added?

Tesla fans lap it up, expecting a pay-off on that big chunk of cash they handed to Tesla and a new toy to show off and impress their friends with. I just wish Elon would be honest about the state of AP and FSD.

And it's most likely that the recent visualization releases that would ideally be a precursor to that same core code coming into mainstream fsd use are actually just the remaining strands of what is now looking like a dead end.

I'm still excited to see where this all goes, but it isn't looking very good in the short-term imo.

 

[diplomat33]

And saying it's "close" to being done is simple misdirection. What is close to being done, version 0.1 Alpha of the re-write code that could never be released to the public? Or V1.0 where everything that the current AP does already works? And how long after that until FSD features can be added?

This. Elon frequently underestimates how difficult it is to finish a feature. So when he says that the rewrite is almost done, he probably means that the version 1 alpha is almost done, meaning they are almost done with the first rewrite but have not really tested it yet. But it will take time to go from an alpha rewrite to actually releasing to the public. So we should not assume that "almost done" means that we will see new features next week or something. When Elon says that new features are coming at the end of this month, take it with a huge grain of salt. He is most likely underestimating again the time it will really take to go from rewrite to a feature that can be deployed to the public.