Autonomous Car Progress

[stopcrazypp]

Waymo has said that it has fewer remote monitors than cars, but has declined to say how many fewer. Obviously the goal will be to have a large ratio. But they never drive the cars remotely, they just solve strategic problems, usually when the car asks.

While expanding to new cities for Waymo is of course not nearly as hard as making it work in the first city, San Francisco was a big step up in difficulty, but they seem very close to having made that step. If they had simply wanted to expand to more of Phoenix, they would have been able to do that in a very short time, but that's not worth doing right now until they are in actual production.

It is worth noting that we're seeing AutoX do unmanned operation in Shenzhen, and not just the suburbs. But their initial suburb (Pingshan) was more difficult than Chandler, AZ but probably not quite as difficult as SF. Waymo doesn't do downtown SF in their service yet but I presume they test in it.

In some sense Waymo has been in SF for many years at some level, but only at large scale recently.

I was going to point out this, but you beat me to it. Waymo has been operating in SF way longer than people are aware (it's not a "new" city for them), just not in the scale they are now. The Bay Area in general they have been in since before they went to AZ (I remember seeing their Firefly vehicle years ago).

Each new city will require some work, though less and less with each one. Only Tesla imagines they can just drive in a new city they have never tested before, or that a person should ever trust their life to a car that's never been verified in that city. Elon keeps saying you will be able to do that this year, for many years including 2022.

For door-to-door L2 (AKA "feature complete" FSD) this is not out of the realm of possibility, in the same way people use AP in roads across the US that Teslas have never driven in before.

 

[voyager]

So, would you say the marketing term is misleading?

The funny thing is I actually don't think it's misleading because FSD Beta is clearly intended to be full self-driving, it just hasn't achieved the safety level necessary to be used without a safety driver yet.

How would geographically determined full self-driving be called? By this I mean:
- in between cities, with traffic driving in the same direction, fully autonomous good enough for you to play games, do some reading, take a nap. IMO, already challenging enough, although some might call this fine-tuned cruise control.
- within urban areas, probably long before you even enter the freeway system around cities, automatic switch (a sort of ILS) to fully manual mode, since assisted drive is too hazardous.

 

[Knightshade]

How would geographically determined full self-driving be called? By this I mean:
- in between cities, with traffic driving in the same direction, fully autonomous good enough for you to play games, do some reading, take a nap. IMO, already challenging enough, although some might call this fine-tuned cruise control.
- within urban areas, probably long before you even enter the freeway system around cities, automatic switch (a sort of ILS) to fully manual mode, since assisted drive is too hazardous.

This is part of what the SAE levels define.

But the really tl;dr verson is:

L2= Car can do PARTS of the dynamic driving task, but not ALL of it- human is still required to be "the driver" legally, and do at least one (or more) pieces of the actual driving task. This is present-day Tesla AP and FSD.

L3= Car can do ALL of the dynamic driving in at least SOME conditions (and those can be defined by a slew of things including geography or road type...or weather, time of day, etc)... but human is required to act as the fallback system if the car thinks it can't keep driving on its own-- this means human is still required in drivers seat-- you can be reading, watching a movie, etc but must be awake and available to take over on short (but not zero) notice. (This is where Mercedes is allegedly going to be with a for-sale car only on highways and only up to like 37 mph, later this year... other german makers had been promising it for years though)

L4= Car can do ALL the dynamic driving in at least SOME conditions -and- a human is -never required- to take over as the fallback- Humans CAN be there if they want- but they can be asleep, sitting in the back, whatever. They're not ever required. (this is where a few RT companies are in very limited deployments/trials in a few cities-- and it's the minimum level the pre-March-2019 FSD buyers were promised their cars would eventually reach)

L5= Car can do ALL the dynamic driving in ALL conditions in which a human can reasonably drive (there certain narrow allowed limits like for legal borders and whatnot). Nobody has this today.

 

[diplomat33]

What we really need to know is - how long does it take Waymo to expand to n+1 city. That will determine whether they can ever be successful.

We only have 2 data points since Waymo has only expanded a ride-hailing service to 2 cities so far but those data points do already show a downward trend.

Waymo started autonomous testing in Chandler: 2017
Waymo deployed driverless in Chandler: 2020
Time: 3 years

Waymo started autonomous testing in SF: 2021
Waymo deployed driverless in SF: likely 2022
Time: 1 year

So 1 year after deploying driverless in Chandler, they were able to expand from testing to driverless in a 2nd city in just 1 year instead of 3 years. That is significant progress. The time it is taking Waymo to expand to n +1 city is decreasing. If the trend continues, eventually the time to expand to n +1 city will be small enough that Waymo can indeed to be successful.

And really it should not be surprising that the time to expand to n +1 city will decrease with each new city. It stands to reason that Waymo is going to learn and improve both their FSD capabilities and their efficiency at deployment with each new city.

So the question is not how long does it take to expand to n +1 city. That number will change over time. The real question is how quickly can Waymo decrease the time to expand to n +1 city to something realistic and scalable. So far, the trend looks promising.

And, Waymo has been doing "mapping + manual testing" in NYC now for 5 months. It will be interesting to see how quickly they get to driverless in NYC once they start actual autonomous testing. That will give us a 3rd data point to see how the trend is going. I predict Waymo will start autonomous testing in NYC this year.

I do think NYC will be another big milestone because once Waymo can do driverless in a dense urban environment like NYC, they will be one step closer to being able to scale in almost any driving environment. So I would expect the time to expand to n + 1 city to decrease even more after they "solve" NYC.

How would geographically determined full self-driving be called? By this I mean:
- in between cities, with traffic driving in the same direction, fully autonomous good enough for you to play games, do some reading, take a nap. IMO, already challenging enough, although some might call this fine-tuned cruise control.
- within urban areas, probably long before you even enter the freeway system around cities, automatic switch (a sort of ILS) to fully manual mode, since assisted drive is too hazardous.

Those would both be L4, just with different ODD attached to them.

Remember that the SAE levels don't include the ODD since it is not realistic to do so. The levels only specify the role of the driver. So for L2-L4, you need to specify both the level and the ODD in order to give a complete description of the automated driving system.

 

[Knightshade]

Waymo started autonomous testing in SF: 2021
Waymo deployed driverless in SF: likely 2022
Time: 1 year

Waymo first tested self driving cars in SF in 2009.

So your math is a bit off.

Waymo self-driving cars spotted in SF

You’ll see these autonomous vehicles waymo than Uber’s.

sf.curbed.com

Story is from 2018, and quotes Waymo themselves saying

San Francisco was one of the first cities where we tested our self-driving cars, dating back to 2009 when we traveled everything from Lombard Street to the Golden Gate Bridge.

here's an actual video- from 2009- of Waymo driving an autonomous car down Lombard street-

So "begin autonomous testing" to "offer driverless rides" took Waymo 13 years in SF.

Far longer than Chandler.


Waymo has only expanded a ride-hailing service to 2 cities so far but those data points do already show an upward trend

 

[diplomat33]

Waymo first tested self driving cars in SF in 2009.

So your math is a bit off.

Waymo self-driving cars spotted in SF

You’ll see these autonomous vehicles waymo than Uber’s.

sf.curbed.com

Story is from 2018, and quotes Waymo themselves saying


here's an actual video- from 2009- of Waymo driving an autonomous car down Lombard street-

So "begin autonomous testing" to "offer driverless rides" took Waymo 13 years in SF.

Far longer than Chandler.

That is all the way back to the early days of testing. Yes, Waymo had a few prototypes in SF way back in 2009. But that was not part of the testing for the current ride-hailing. I am counting when Waymo started actual testing at scale of their ride-hailing in SF which was in 2021.

 

[Knightshade]

That is all the way back to the early days of testing. I am counting when Waymo started actual testing at scale in SF which was in 2021.

We call this moving goalposts


I mean, Tesla only began city streets testing "at scale" less than a year ago- so I guess we can say they're not behind schedule after all right?

No, of course not.

When they started is when they started. If it took em 12 years to go from "small testing" to "testing at scale" in a city we can't just ignore it took them 12 years.

They took much less time to do that in Chandler- but you'd expect them to as it's a vastly simpler place to do it.

SF is much harder, and it took them much longer- exactly as you'd expect.

Trying to spin that into LOOK HOW FAST THEY GOT SF WORKING is... not accurate.

 

[OncomingStorm]

just my opinion but I really think its going to be much harder than people think for these companies to scale up. the big problem with their approach is that driving in one city is VERY different than driving in another. everything, local drive styles, lane lines, signs, laws, parking habits, literally everything lol changes from one place to another. hell even for a human it can be a nightmare to go to somewhere they've never been before so to think that because your AI or whatever preforms well in one very specific place that it will perform just as well as another is a fallacy. each time it will have to be retrained on a pretty significant scale. sure some things translate, but if you drive like you are in FL in MI, you are likely to kill someone lol, so they almost have to remove a lot of the "learning" their system has done each time and start from scratch.

obvious speculation on my part. and i could be wrong, but that kind of piecemeal approach is too basic to even be called self driving imho. thats more akin to a hotwheels track, yes my little cars go round and round but they will never be able to do anything more and will fail at the drop of a hat when encountering anything outside of their very specific programming.

 

[powertoold]

Until recently, many here thought that Mobileye was ahead of Tesla.

Lol.

 

[OncomingStorm]

Until recently, many here thought that Mobileye was ahead of Tesla.

Lol.

haha yeah, the linchpin for me is if these systems are so much better than Tesla's, why are they not produced in any kind of significant way? if you had the plans for a product that would literally make you billions, do you really think supply chain issues would stop you from producing it in a large way? people need to give less credence to marketing material and look at what these companies are actually doing.

 

[diplomat33]

Until recently, many here thought that Mobileye was ahead of Tesla.

Lol.

Mobileye is ahead of Tesla. They have vision-only that can do door to door and are already testing L4 robotaxis in several cities. Tesla is not.

 

[diplomat33]

We call this moving goalposts


I mean, Tesla only began city streets testing "at scale" less than a year ago- so I guess we can say they're not behind schedule after all right?

No, of course not.

When they started is when they started. If it took em 12 years to go from "small testing" to "testing at scale" in a city we can't just ignore it took them 12 years.

They took much less time to do that in Chandler- but you'd expect them to as it's a vastly simpler place to do it.

SF is much harder, and it took them much longer- exactly as you'd expect.

Trying to spin that into LOOK HOW FAST THEY GOT SF WORKING is... not accurate.

Well, if you want to get super technical and start with the very first prototype testing in SF, sure you would be right. You want to go back to the first prototype testing in SF because it fits your agenda of claiming Waymo took longer to deploy in SF. But that is silly. That prototype in 2009 was completely different from the ride-hailing that they are deploying now now in SF. I think it makes more sense to start with when Waymo actually started testing the ride-hailing that they plan to deploy to see how long it takes to test and deploy a new service, not start with the first prototype.

And I am consistent with Tesla too. I would say Tesla started testing FSD Beta 1 year ago because that is when they released FSD Beta to testers. Obviously, Tesla started testing some earlier versions of FSD back in 2016. But I would not go back to AP1 and say Tesla took 8 years to test FSD Beta. That would be silly.

 

[Daniel in SD]

haha yeah, the linchpin for me is if these systems are so much better than Tesla's, why are they not produced in any kind of significant way? if you had the plans for a product that would literally make you billions, do you really think supply chain issues would stop you from producing it in a large way? people need to give less credence to marketing material and look at what these companies are actually doing.

The question is how much money could they make off of a product like FSD Beta? I think the market for supervised self-driving is smaller than you think and there's all the bad press that will come from the inevitable accidents caused by automation complacency. If FSD Beta is ever available to everyone it would be interesting to know how many people actually use it regularly.

 

[diplomat33]

haha yeah, the linchpin for me is if these systems are so much better than Tesla's, why are they not produced in any kind of significant way? if you had the plans for a product that would literally make you billions, do you really think supply chain issues would stop you from producing it in a large way? people need to give less credence to marketing material and look at what these companies are actually doing.

The difference is Tesla is willing to sell a "beta" product, relying on driver supervision to hopefully avoid too many accidents. Other companies like Mobileye and Waymo are more responsible and they only deploy products to the public once the products are fully validated and tested to be safe.

 

[diplomat33]

Some of you make fun of Waymo for the sensors on the roof. Waymo is absolutely stylish compared to the Apple autonomous car. Yikes. Could Apple have made the roof sensors any bigger? It looks like a king size mattress tied to the roof. LOL.

 

[Knightshade]

Some of you make fun of Waymo for the sensors on the roof. Waymo is absolutely stylish compared to the Apple autonomous car. Yikes. Could Apple have made the roof sensors any bigger? It looks like a king size mattress tied to the roof. LOL.

That's the Apple Car Pro Max, the sensors on the Apple Car SE will be much smaller

 

[OncomingStorm]

The difference is Tesla is willing to sell a "beta" product, relying on driver supervision to hopefully avoid too many accidents. Other companies like Mobileye and Waymo are more responsible and they only deploy products to the public once the products are fully validated and tested to be safe.

ignoring FSD Beta, which obviously requires addtional steps to get, i dont feel like any of the features that are labeled beta are dangerous in anyway, quite the contrary, half the reason people complain about AP is because it's overly safe / reactive to threats. seems to me that if they really didn't care about safety they would just release everything with 0 testing to everyone. obvisouly thats rediculus, much like the claim that tesla is releaseing "unsafe" features.

 

[EVNow]

The difference is Tesla is willing to sell a "beta" product, relying on driver supervision to hopefully avoid too many accidents. Other companies like Mobileye and Waymo are more responsible and they only deploy products to the public once the products are fully validated and tested to be safe.

The difference is people are willing to buy a beta product from Tesla.

Actually even Amazon sells "day 1" products by invitation only - what are essentially beta products and there is usually a big waiting list to buy them.

Infact there are people willing to just buy various parts and assemble stuff on their own !

Turns out there are enough people in the world who haven't lost their curiosity. Thank God for that.

https://www.amazon.com/Introducing-Amazon-Astro/dp/B078NSDFSB

 

[S4WRXTTCS]

This claim is.... exaggerated. By quite a lot.

AFAIK the only place anything like what you claim has happened is in ONE country (Germany).

And it only happened in ONE regional court.

And it was over the term autopilot, not FSD, so the court clearly got the whole ruling wrong.

And AFAIK led to no actual action, and AFAIK Tesla, rightly, continues using that term even in the EU.

Here is what I was referring to

German court rules that Tesla misled consumers on Autopilot and Full Self Driving

Tesla mislead consumers on the abilities of its automated driving systems, a Munich court ruled on Tuesday.

www.cnbc.com

 

[Knightshade]

Here is what I was referring to

German court rules that Tesla misled consumers on Autopilot and Full Self Driving

Tesla mislead consumers on the abilities of its automated driving systems, a Munich court ruled on Tuesday.

www.cnbc.com

Yes, and your claims were...not accurate... as I laid out about that specific ruling in some detail.

It only happened in one country, not "countries" as you claimed.

It only happened in one regional (not national) court in that country.

It was about a court not understanding what the actual word autopilot means and instead imagining it "misleads" you into thinking it requires no driver- despite the fact the term has been used to describe an aircraft system for decades that still requires a pilot

And to my knowledge has resulted in no actual change in Teslas behavior in that country, as they still sell autopilot in Germany with the same wording as before the court case.

In fact- just as a quick sanity check on myself, I VPNed into Germany, went to Teslas website, changed my country to Germany there, and every word the court did not want Tesla to use is still being used 2 years later

Suggesting the verdict held no actual force of law (or was overturned by a higher court but I can't find a record of that)


So to sum up-- One regional court in one place misunderstood what the word meant and issued a bad ruling that didn't actually have any effect. Which is pretty far from your claim that multiple countries with consumer protection laws have used them to cause...anything at all... regarding AP or FSD.