Autonomous Car Progress

[Knightshade]

Really it depends on what you're trying to do.

I mean, Waymo probably has more real world data on Chandler Arizona than even Tesla does- so as long as they only want to keep offering service there they're good to go.

Scaling though- not so much.

 

[Bladerskb]

Really it depends on what you're trying to do.

I mean, Waymo probably has more real world data on Chandler Arizona than even Tesla does- so as long as they only want to keep offering service there they're good to go.

Scaling though- not so much.

You mean there isn't 100-200 Teslas in Phoenix? Do Waymo also have more cars in SF than Tesla does?
Is that why Waymo can go tens of thousands of miles without a safety disengagement and Tesla is struggling to break the 10 miles barrier?

 

[Knightshade]

You mean there isn't 100-200 Teslas in Phoenix?

By now I'm sure there are.

But Waymos been working on this for longer than Tesla has been mass producing cars. Years longer.

And still haven't managed to launch consumer-facing taxis outside of that one tiny suburb.

Do Waymo also have more cars in SF than Tesla does?

No, but then they don't offer consumer RT service in SF either- despite again having spent many years "gathering data" there.

Almost like scaling beyond a single city takes Waymo many years and they still haven't got it right or something.

Like their approach doesn't scale well or something.

WHO COULD HAVE GUESSED.

Is that why Waymo can go tens of thousands of miles without a safety disengagement and Tesla is struggling to break the 10 miles barrier?

... on what data are you basing this?

AFAIK only in CA do we have objective data of disengagements of systems running above L2, and apart from a tiny # of miles in 2019 driven at L3 for the Autonomy day video, Tesla has no data in that set to compare with Waymo in the last ~5 years.

(in that 2019 report BTW Tesla reported zero disengagements at all- though again the miles driven were very very low)

 

[Daniel in SD]

... on what data are you basing this?

AFAIK only in CA do we have objective data of disengagements of systems running above L2, and apart from a tiny # of miles in 2019 driven at L3 for the Autonomy day video, Tesla has no data in that set to compare with Waymo in the last ~5 years.

Yup, people have to remember that what we're seeing in these videos isn't the latest build. Elon is always testing something far more advanced than what is released to the public. I expect V10 will be a step change and our minds will be blown.

 

[diplomat33]

No one believes Waymo thinks they can do it all based on simulation.

But that is what "joe" claimed:

"Why bother with labeling Andrej when Waymo says you can get all the data you want from simulations? "

He did not say "some", he said "all"! How else can you read what he wrote?

 

[ZeApelido]

But that is what "joe" claimed:

"Why bother with labeling Andrej when Waymo says you can get all the data you want from simulations? "

He did not say "some", he said "all"! How else can you read what he wrote?

Because that Joe is me and I was just being overly sarcastic

 

[diplomat33]

Almost like scaling beyond a single city takes Waymo many years and they still haven't got it right or something.

Like their approach doesn't scale well or something.

WHO COULD HAVE GUESSED.

The problem is not scaling. If it was just a matter of deploying a lot of test vehicles, Waymo could scale up now if they wanted to. But that is not Waymo's goal. Waymo's goal is to deploy safe and reliable robotaxis with no human driver. So Waymo wants the FSD to be good enough to remove the safety driver BEFORE they scale up. So Waymo needs to improve the safety and reliability first. Obviously, solving FSD to where you don't need a safety driver at all, is very difficult. That is why Waymo has not deployed robotaxis in more areas yet. But once they get the FSD to be "good enough" for their safety metrics, they will scale quickly.

Tesla does not care about "FSD" being good enough since the driver is required to supervise. So Tesla can scale first even when "FSD" is not ready yet.

Basically, Tesla is scaling first and then trying to solve FSD. Waymo is solving FSD first and then they will scale.

 

[mikes_fsd]

But that is what "joe" claimed:

"Why bother with labeling Andrej when Waymo says you can get all the data you want from simulations? "

He did not say "some", he said "all"! How else can you read what he wrote?

Congrats, you're inability to focus on the actual topic, has detailed this thread to discussing some bs about simulation!

 

[diplomat33]

Congrats, you're inability to focus on the actual topic, has detailed this thread to discussing some bs about simulation!

You are the one that started the topic by posting the BS tweet from "joe".

 

[mikes_fsd]

You are the one that started the topic by posting the BS tweet from "joe".

I took an effing screenshot just so you wouldn't jump to the wrong conclusions, but no!

And, my entire focus was on what Karpathy said not on what some random account on Twitter said

 

[Doggydogworld]

So Waymo wants the FSD to be good enough to remove the safety driver BEFORE they scale up. So Waymo needs to improve the safety and reliability first. Obviously, solving FSD to where you don't need a safety driver at all, is very difficult. That is why Waymo has not deployed robotaxis in more areas yet. But once they get the FSD to be "good enough" for their safety metrics, they will scale quickly.

Waymo is obviously safe enough to remove safety drivers, since they already did it. There's nothing unique about the current 50 square miles that they couldn't replicate in the 3000 other square miles of suburban Phoenix or hundreds of thousands of square miles in other good-weather suburbs.

The problem is the business model. Suburban Robotaxis lose money. Lots of money. Going from 50 to 50,000 square miles only multiplies the losses by 1000. Even Google doesn't have the stomach for that.

A start-up entrepreneur would have iterated the business model a couple dozen times by now until finding one that works. Campuses, retirement villages, business shuttles, airports, U-drive, whatever. A fake-it-til-you-make-it huckster would have scattered services around 20 cities and raised 50 billion. But Waymo is not led by entreprenuers, so they plowed straight ahead with a bad business model.

They are finally pivoting to urban and freight, both business models that work. But they've mostly squandered their technical lead.

 

[Knightshade]

The problem is not scaling

Clearly it is though.

Otherwise they'd have scaled.

. If it was just a matter of deploying a lot of test vehicles, Waymo could scale up now if they wanted to.

No, they couldn't- because it takes years and years to get the mapping done.

See all the cities they "began" mapping years ago and still have no service available and are still sending mapping cars around.

But that is not Waymo's goal. Waymo's goal is to deploy safe and reliable robotaxis with no human driver. So Waymo wants the FSD to be good enough to remove the safety driver BEFORE they scale up.

But they already did that.

In AZ.

Almost 2 years ago.

Waymo Is Letting People Choose Rides in Completely Driverless Cars

Early riders on Waymo's customer list are getting the choice to be the only occupants of self-driving cars.

www.caranddriver.com

And then scaled... nowhere.

Waymo is obviously safe enough to remove safety drivers, since they already did it.

Yup.

There's nothing unique about the current 50 square miles that they couldn't replicate in the 3000 other square miles of suburban Phoenix or hundreds of thousands of square miles in other good-weather suburbs.

Well, there is.

The super precise super HD maps that apparently takes years to develop for each place and then constantly need to be KEPT updated.

They announced starting to map LA, for example, back in fall 2019... years after they began mapping San Francisco and other cities where they also haven't gotten far enough to actually offer service yet

 

[diplomat33]

The problem is the business model. Suburban Robotaxis lose money. Lots of money. Going from 50 to 50,000 square miles only multiplies the losses by 1000. Even Google doesn't have the stomach for that.

A start-up entrepreneur would have iterated the business model a couple dozen times by now until finding one that works. Campuses, retirement villages, business shuttles, airports, U-drive, whatever. A fake-it-til-you-make-it huckster would have scattered services around 20 cities and raised 50 billion. But Waymo is not led by entreprenuers, so they plowed straight ahead with a bad business model.

I don't completely disagree about the bad business model. I think Waymo launched their first ride-hailing in Chandler because they thought it would be relatively easy and saw it as a "proof of concept" that would let the public experience a real robotaxi service. But as you said, Chandler's population density is too low to make a robotaxi service profitable.

But hindsight is 20/20. It's easy to look back and say Waymo should have done things differently. But back in 2017, when Waymo started the service in Chandler, the FSD was not as good as it is today and the costs were a lot higher than they are now. So at the time, it probably did not look like they could expand yet. Waymo did what they thought was logical at the time: get the FSD good enough to remove the safety driver and reduce costs.

But they've mostly squandered their technical lead.

Somewhat maybe. Certainly, the competition is not sitting on their laurels. But Waymo still has the best FSD software and hardware. I say that because Waymo is a top innovator in ML. Waymo showed NN that is better than the current state of the art. Waymo has the best disengagement rate in CA. Waymo also custom builds all their sensors and they are among the best in the industry. And nobody in the US has duplicated the Waymo service in Chandler yet. So for all the talk about how Waymo can't get out of a 50 sq mi area, the fact is that nobody else has even been able to do that yet.

But I think Waymo needs to make a big move very soon. Cruise has a disengagement rate that is almost as good as Waymo in CA. Cruise is also poised to launch a ride-hailing service in SF probably next year. Baidu is launching a ride-hailing service in China. Zoox is starting production on their robotaxi. If Waymo does not make a big move and just continues to try to make their FSD perfect in a geofenced area, then I think they will lose their lead.

 

[diplomat33]

No, they couldn't- because it takes years and years to get the mapping done.

See all the cities they "began" mapping years ago and still have no service available and are still sending mapping cars around.

Wrong. It does not take years to map cities. You are confusing "not launching a service" with "still mapping". They finished the mapping a long time ago and deployed autonomous vehicles, just not a public service. They are not still sending mapping cars to those cities. They have test cars driving autonomously. Not the same as mapping cars.

But they already did that.

In AZ.

Almost 2 years ago.

Waymo Is Letting People Choose Rides in Completely Driverless Cars

Early riders on Waymo's customer list are getting the choice to be the only occupants of self-driving cars.

www.caranddriver.com

And then scaled... nowhere.

Because it takes time to "solve FSD" and make sure FSD is safe enough. Waymo is not going to just deploy their service in new areas before making sure it is safe enough in that new area. Waymo does not deploy untested "beta" software to the public like Tesla does. Waymo actually makes sure their software is reliable and safe before letting the public ride.

I don't see Tesla deploying any driverless services anywhere. Tesla has been working on FSD for 5 years and still has not deployed any driverless service. At least, Waymo has a driverless service somewhere and has the ability to scale to other cities when they are ready. Tesla has not shown that they can even do any driverless yet.

 

[Zhelko Dimic]

He goes on to say a fully autonomy vehicle that is much safer than a person "is much easier than you would think"


So it sure does not sound like he IDed it as a difficult problem back then.

Quite the opposite.

I used to be mad about Elon underestimating problems.

I've recently realized that's one of his greatest strength, when combined with ability and willingness to change course.

I.e. he's just stubborn enough to persist on hard problems that could be solved in "reasonable" time. The greater reward, the longer "reasonable" time is...

 

[Daniel in SD]

I'm not sure what the obsession with Waymo and mapping is. Look at their safety data, have any of their collisions been due to mapping? It seems like the least of their problems. It looks like their biggest problem is other road users.

https://storage.googleapis.com/sdc-prod/v1/safety-report/Waymo-Public-Road-Safety-Performance-Data.pdf

Maybe they haven't scaled because they haven't proven safety greater than a human? Or maybe because they're using so much remote assistance that they'd just lose more money expanding to a new city.

I used to be mad about Elon underestimating problems.

I've recently realized that's one of his greatest strength, when combined with ability and willingness to change course.

I.e. he's just stubborn enough to persist on hard problems that could be solved in "reasonable" time. The greater reward, the longer "reasonable" time is...

I think 99% of the people here who have a beef with Elon are only upset because Tesla took their money for FSD years ago and will probably never deliver what was promised. No one is going to be upset if he doesn't meet his timeline for colonization of Mars (unless he starts charging for tickets).

 

[stopcrazypp]

I took an effing screenshot just so you wouldn't jump to the wrong conclusions, but no!

And, my entire focus was on what Karpathy said not on what some random account on Twitter said

Unfortunately, when you posted the link, the forum software included that "joe's" comment. That's what I hate about the auto-embed feature here. It adds things that you didn't necessarily want to include (sometimes the goal is just to post a link).

 

[daniel]

Waymo is obviously safe enough to remove safety drivers, since they already did it. There's nothing unique about the current 50 square miles that they couldn't replicate in the 3000 other square miles of suburban Phoenix or hundreds of thousands of square miles in other good-weather suburbs.

The problem is the business model. Suburban Robotaxis lose money. Lots of money. Going from 50 to 50,000 square miles only multiplies the losses by 1000. Even Google doesn't have the stomach for that.

...

Robotaxis lose money because they are so expensive. What robotaxis are good for is as a real-world development platform. I think that Waymo has not scaled up to more cities because a development platform does not need more cities. It does need more variety of terrain and conditions, but robotaxis cannot be deployed until they are safe in the terrain and conditions where they will operate. So expansion to more and more difficult conditions will only proceed as fast as the platform can advance to handle those conditions.

The near-term goal is twofold: 1. Develop hardware and software capable of operating driverless in any conditions where a human could drive; and 2. Bring down the cost of the car enough that it becomes affordable/economical.

There's no reason to expand beyond Chandler until the car can handle conditions more difficult than Chandler presents. This is turning out to be difficult. And a driverless car won't be available to consumers until the cost can be brought down enough that they can sell enough of them to be profitable.

 

[mikes_fsd]

This was discussed at several different times in the past.

I think this is the first confirmation from Elon that the mic will be used to listen to surroundings.

 

[diplomat33]

Some of you might find this interesting. According to a study done in 2019 entitled "How Safe Is Safe Enough for Self-Driving Vehicles?" by Liu et al., the tolerable risk is 3-4x safer than humans, the broadly acceptable risk is 100x safer than humans. So that gives us a pretty good idea of the range of AV safety that the public could accept.

Abstract
Self-driving vehicles (SDVs) promise to considerably reduce traffic crashes. One pressing concern facing the public, automakers, and governments is “How safe is safe enough for SDVs?” To answer this question, a new expressed-preference approach was proposed for the first time to determine the socially acceptable risk of SDVs. In our between-subject survey (N = 499), we determined the respondents’ risk-acceptance rate of scenarios with varying traffic-risk frequencies to examine the logarithmic relationships between the traffic-risk frequency and risk-acceptance rate. Logarithmic regression models of SDVs were compared to those of human-driven vehicles (HDVs); the results showed that SDVs were required to be safer than HDVs. Given the same traffic-risk-acceptance rates for SDVs and HDVs, their associated acceptable risk frequencies of SDVs and HDVs were predicted and compared. Two risk-acceptance criteria emerged: the tolerable risk criterion, which indicates that SDVs should be four to five times as safe as HDVs, and the broadly acceptable risk criterion, which suggests that half of the respondents hoped that the traffic risk of SDVs would be two orders of magnitude lower than the current estimated traffic risk. The approach and these results could provide insights for government regulatory authorities for establishing clear safety requirements for SDVs."

Source:

How Safe Is Safe Enough for Self‐Driving Vehicles?

Self-driving vehicles (SDVs) promise to considerably reduce traffic crashes. One pressing concern facing the public, automakers, and governments is “How safe is safe enough for SDVs?” To answer this ...

onlinelibrary.wiley.com