EVNow
Well-Known Member
Only thing I’d add is : “I do think Waymo's FSD is good enough to work in my area now” - if Waymo had spent the same time it did in Chandler and when it is not raining or snowing.
Last edited:
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Autonomous Car Progress
- Thread starter stopcrazypp
- Start date
-
- Tags
- Autonomous Vehicles
diplomat33
Average guy who loves autonomous vehicles
EVNow
Well-Known Member
Knightshade
Well-Known Member
diplomat33
Average guy who loves autonomous vehicles
So you're saying they have a scaling problem
Nice try but no, I am not saying that at all. I am just saying that I don't think Waymo will bother deploying robotaxis in a town of 60k in rural Indiana or if they do, it will be after they deploy in major cities first. But I am confident that Waymo will scale when the Waymo Driver is good enough to scale. I just think they will start with major cities where they can maximize profit.
diplomat33
Average guy who loves autonomous vehicles
As I have said before, Waymo does not have a scaling problem. But the issue with deploying driverless robotaxis is that your robotaxis have to be able to drive safely without any human intervention in all ODDs that you want to deploy in. You need the FSD to be 99.99999% reliable before you can scale. It is a much higher standard than what Tesla is doing because Tesla can rely on driver supervision to pick up the slack if FSD Beta can't handle something. Waymo robotaxis have to be good enough to drive safely without driver supervision. But once the Waymo Driver is good enough everywhere, I don't see anything that would stop Waymo from scaling.
Last edited:
Knightshade
Well-Known Member
Dude who is making the video is a well known FUD spreader- and weirdly the reporter failed to mention that guy also has heavy investments in LIDAR companies too.
Can't IMAGINE why he's want to spread FUD about Teslas system that doesn't use LIDAR.
Nope....can't think of any reason....
EVNow
Well-Known Member
A lot of inherent conflicts there.
First you state "Waymo does not have a scaling problem". At least say it is your opinion ... since definitely all indications are they are unable to scale. It is definitely not a "fact".
Then, you say the problem is they need to be have seven 9s accuracy. Their disengagement report actually indicates more like 99.9% i.e. three 9s. Anyway, I'll let it pass.
Then you say - "But once the Waymo Driver is good enough everywhere, I don't see anything that would stop Waymo from scaling.". Thats a headscratcher - if they are good everywhere - they are good everywhere ... what ?!
Let me say "scaling" in another way. Unlike Tesla, Waymo is "geohacked" to certain blocks (sorry, can't say cities). They are only "good enough" in those areas. They are not good in other areas. Waymo has a big problem making their cars "good enough" in more areas a.k.a. geoscaling problem.
EVNow
Well-Known Member
diplomat33
Average guy who loves autonomous vehicles
Yes, it is my opinion. But I think it is an informed opinion. I have studied how Waymo's FSD works. The perception, prediction and planning stacks are generalized so they can work everywhere. Building and updating HD maps is more and more automated so the process is getting faster and cheaper. So I don't think HD maps will prevent scaling. Cost of sensors like lidar is getting cheaper. Cost of remote assistance will go down as Waymo adds more 9's of reliability. Waymo is transitioning to EVs so the maintenance cost of the fleet will go down. So I don't think costs will prevent scaling. And Waymo has already developed an experienced autonomous driver that can handle parking lots, highways, suburban streets and some dense urban streets. Obviously, Waymo still has work to do. They still need to solve dense urban driving, hence why they are testing in NYC. So IMO, the "march of 9's" is the only thing really preventing Waymo from scaling. If I am right then once Waymo achieves 99.99999%, they should be able to scale. In fact, I would suggest that the march of 9's is the only major thing preventing everybody (working on AVs) from scaling. You can do like Tesla and scale before you achieve 99.99999% if you deploy a driver assist but if you want to deploy driverless cars, you must achieve 99.99999%.
Yes, I am saying that Waymo does not have enough 9's yet. That is why they can't scale yet.
I think my position is very logical:
Assumption: You need 99.99999% reliability to scale robotaxis.
Fact: Waymo is not there yet.
Deduction: Waymo can't scale yet. When Waymo does get there, they will be able to scale.
I am simply saying that when the Waymo Driver works well enough everywhere, ie has enough 9's, then Waymo can go ahead and deploy robotaxi services everywhere without driver supervision.
bradtem
Robocar consultant
Waymo certainly scales better than Tesla does now. Should Tesla get a computer vision breakthrough at some unknown date, it will scale well, but today as an FSD system (rather than driver assist) it does not scale at all, since it does not work and its only path to working requires a breakthrough. Well, its only path to working based on their current strategy. If they switched to the "Tesla Master Plan" strategy (start with expensive, get cheaper later) and used maps, lidar and imaging radar, they would have a path to working, and the "get cheaper later" would be their path to scaling.
EVNow
Well-Known Member
I'm not sure everyone is using "scaling" in the same way, leading to confusion.
Let me repeat what I've been saying for 3 years.
When it comes to autonomous driving there are multiple dimensions. But, lets concentrate on 2 items.
- Geography
- Quality / Failure Rate
Being good on one dimension doesn't say much about the other dimension. Do not conflate and confuse the dimensions. This is a "vector" - not a scalar that can be easily compared. A value of a vector in one dimension cannot be compared to value of the vector in another dimension.
Tesla is "9" or "great" when it comes to Geography dimension. That says nothing about the other dimension - Quality where it is "1" or "poor".
Similarly, Waymo is "9" or "great" when it comes to Quality dimension. That says nothing about the other dimension - Geography where it is "1" or "poor".
So, it makes zero sense to claim Tesla scales well in Quality (i.e. "I'll be shocked if we don't surpass human level driving this year") given where it is and has been for the past year. Similarly, it makes zero sense to claim Waymo "certainly scales better than Tesla" when it comes to geography given where it is today and where it has been in the past 4 years. The pace of development in either case does not lend to optimism for scaling quickly in the future.
diplomat33
Average guy who loves autonomous vehicles
I am defining "scaling" as deploying a driverless robotaxi service to the public in a given area. So right now, Waymo has scaled to 1.8 areas. They have driverless in one area (Chandler) and almost driverless in another area (SF). But I am not making any claims that Waymo will scale faster because they rate a 9 in Quality. I am simply saying that the march of 9's is a prerequisite for scaling because you need high reliability before you can deploy a driverless robotaxi service. So I am suggesting that as Waymo adds more 9's, they will be able to scale out (ie deploy more robotaxis in more areas).
EVNow
Well-Known Member
So you are saying, magically, if Waymo drives better in Chandler they will drive better everywhere ? That would be true of Tesla, obviously not Waymo. Apparently Waymo drives well in some parts of Chandler but not in rest of Phoenix.
Scaling generally says how fast you can expand. Currently Waymo has done 1.5 cities in 4 years. Very poor scaling.
diplomat33
Average guy who loves autonomous vehicles
No. I am not saying that at all. I am saying that as Waymo learns to drive in new areas, they will eventually be able to drive everywhere. That's why Waymo is testing in different areas like SF, LA and now in NYC. Waymo is gaining driving experience in different environments to expand their driving beyond Chandler.
It's the first 4 years at the beginning of developing FSD. It is still relatively early in the development. Exponential curves start very slow and flat in the beginning.
If we measure scaling by deployment of robotaxis, Tesla's scaling is 0 in 6 years since they have not deployed any driverless robotaxis or AVs. So Tesla's scaling of robotaxis is even worse than Waymo, In fact, Waymo is currently scaling robotaxi deployments faster than Tesla!
And considering that nobody has solved FSD yet and deploying robotaxis is completely new territory for everybody, I don't think anybody knows what a good rate of scaling robotaxis should be. So complaining that Waymo is not scaling fast enough at something that nobody has done before, is silly.
Last edited:
bradtem
Robocar consultant
But what basis do you have for the claim that Tesla FSD is a "9" in geography? Indeed what basis do you have for it being more than a zero? Is there any geography in which it shows signs of working? Working means "taking the driver out" with full self-driving. Note that even with Waymo, AutoX, Nuro and others who have taken the driver out, that still has not meant they were ready, it perhaps means they are a few years form being ready. Where is Tesla on that scale.
"Doesn't work everywhere" isn't a sign of scalability. What is more correct is that Tesla is *trying* to work everywhere, but then so are several others. You don't think Waymo could try to work everywhere? And probably do it less poorly than Tesla? What grounds do you have for that belief? MobilEye hasn't taken their driver out yet, but what they did do was demonstrate they could drive in a whole new city with just 2 weeks of work by 2 non-engineering staff. I would say that's some evidence of scalability.
You do understand the the tech that Tesla uses they did not invent -- many elements of it were invented by people at Google, who share their stuff with Waymo.
They are not independent elements at all. Quality has a threshold to it. If you aren't near that threshold, it doesn't matter how many places you are also not near that threshold. I mean sure, if there were evidence that Tesla was not very, very, very far from the threshold you could talk about the potential of them making their breakthrough. But breakthroughs, being what they are, are not predictable.
Not saying Tesla doesn't have some assets and some innovating things they are doing. But you have to understand where they sit in the hierarchy of teams, and it doesn't matter how many places they drive unacceptably badly. All the others could also drive unacceptably badly on all roads should they want to follow that path, but they think it a foolish path.
Why is the Tesla Master Plan of start expensive and low volume and then later go cheap and high volume, wrong? The Roadster and Model S were niche products with limited markets that could never scale the way the Model 3 does, or how the upcoming Model 2 will. Tesla would do well to pay attention to the strategy behind those cars.
"Doesn't work everywhere" isn't a sign of scalability. What is more correct is that Tesla is *trying* to work everywhere, but then so are several others. You don't think Waymo could try to work everywhere? And probably do it less poorly than Tesla? What grounds do you have for that belief? MobilEye hasn't taken their driver out yet, but what they did do was demonstrate they could drive in a whole new city with just 2 weeks of work by 2 non-engineering staff. I would say that's some evidence of scalability.
You do understand the the tech that Tesla uses they did not invent -- many elements of it were invented by people at Google, who share their stuff with Waymo.
They are not independent elements at all. Quality has a threshold to it. If you aren't near that threshold, it doesn't matter how many places you are also not near that threshold. I mean sure, if there were evidence that Tesla was not very, very, very far from the threshold you could talk about the potential of them making their breakthrough. But breakthroughs, being what they are, are not predictable.
Not saying Tesla doesn't have some assets and some innovating things they are doing. But you have to understand where they sit in the hierarchy of teams, and it doesn't matter how many places they drive unacceptably badly. All the others could also drive unacceptably badly on all roads should they want to follow that path, but they think it a foolish path.
Why is the Tesla Master Plan of start expensive and low volume and then later go cheap and high volume, wrong? The Roadster and Model S were niche products with limited markets that could never scale the way the Model 3 does, or how the upcoming Model 2 will. Tesla would do well to pay attention to the strategy behind those cars.
voyager
Active Member
voyager
Active Member
Seems we have come full circle within this thread as far as Cruise concerned. It took four years.
voyager
Active Member
The posting I was referring to above, dates back from Oct. 9, 2017.
Cruise' experimenting in SF is interesting. Not the easiest infrastructure there.
Then again, the robo-taxis are performing services only between 11 PM and 5 AM.
So, no school kids on the road to avoid.
Cruise' experimenting in SF is interesting. Not the easiest infrastructure there.
Then again, the robo-taxis are performing services only between 11 PM and 5 AM.
So, no school kids on the road to avoid.
Similar threads
- Article
