Waymo

[diplomat33]

Waymo testing in Austin!! Could be sign Austin will be next city after LA.

https://twitter.com/x/status/1641758578363998208

 

[diplomat33]

According to the Waymo blog, Waymo is testing "across central and East Austin, including downtown, Rainey Street, Clarksville, Bouldin Creek, the Market District, Holly, and the Capitol."

So it looks like a pretty good size geofence that includes the dense urban parts of Austin.

 

[diplomat33]

https://twitter.com/x/status/1644817600390496262

 

[Watts_Up]

Put Tesla FSD beta with no human driver in the same geofence as Waymo I believe Tesla FSD beta would do worse than Waymo.

In terms of outside the geofence, the Waymo Driver can work everywhere .

Both FSD and Driver 5 have limitations so it is difficult to make complete comparisons.
So are you making your assumptions and judgments or are you making such comments based on reported facts.

I was surprised for example, in the following video, to learn that waymo could not use a freeway:

 

[diplomat33]

Both FSD and Driver 5 have limitations so it is difficult to make complete comparisons.
So are you making your assumptions and judgments or are you making such comments based on reported facts.

My statement is an educated opinion. I know how Waymo uses generalized NN for perception, prediction and planning. Since the NN are generalized, we can extrapolate how they might work outside the geofence. But obviously, we can't know for a fact how Waymo actually does outside the geofence since there are no videos of the Waymo Driver outside the geofence.

You cannot make any fair comparison between Tesla's L2 and Waymo's L4 based on a 15 minute video. That is silly. The video will only cover a small number of cases. It won't cover a lot of situations. So all it proves is that in this particular drive, under those specific circumstances, Tesla could do the drive with zero interventions. But zero interventions is not the same as driverless. It does not mean that Tesla can do driverless. Remember that Tesla FSD is L2 while Waymo is L4.

I was surprised for example, in the following video, to learn that waymo could not use a freeway:

To be clear, Waymo can use freeways, it just does not let the public take driverless rides on freeways yet, There is a difference. That is because liability is much higher when doing driverless. So Waymo is restricting the public on freeways until they finish validation. Tesla FSD does freeways but it has a driver. So Tesla is only doing L2 freeway.

But here is proof Waymo can do freeways. The Waymo CEO recently did a driverless ride on freeways:

https://twitter.com/x/status/1620144265572814855

Waymo is testing and validating on freeways. Once they finish validation, they will allow the public to take driverless rides on the freeway. And then we will see public videos of Waymo on the freeway and it will be L4 freeway. My guess is that it will happen this year.

 

[Doggydogworld]

But here is proof Waymo can do freeways. The Waymo CEO recently did a driverless ride on freeways:

https://twitter.com/x/status/1620144265572814855

Where does it say this was driverless?

 

[ThomasD]

What about Texas Toll Road. SH 130

A Trip Down the Fastest Highway in America

The newly opened stretch of State Highway 130 from Austin to Seguin has an 85 mph speed limit, the fastest in the country. Beyond the debate over whether it's safe, what's it like to drive?

www.texastribune.org

 

[diplomat33]

Where does it say this was driverless?

The tweet did not explicitly state it so I probably presumed a bit by using the word "driverless". But whether there was a safety driver or not, it was still a L4 drive. So it still proves that Waymo can do L4 highways.

What about Texas Toll Road. SH 130

A Trip Down the Fastest Highway in America

The newly opened stretch of State Highway 130 from Austin to Seguin has an 85 mph speed limit, the fastest in the country. Beyond the debate over whether it's safe, what's it like to drive?

www.texastribune.org

What about it? Are you asking if autonomous vehicles can do 85 mph? Sure, why not.

 

[diplomat33]

In this episode, we cover:
(00:00) Shweta’s background
(03:47) What Shweta and her team are responsible for at Waymo
(05:30) About the autonomous driving vehicle hardware, software, and simulation tools
(08:14) Differences in working at Waymo vs. a more traditional software company
(11:02) How Waymo builds trust with riders and the difference between driver assist and fully autonomous
(13:57) An example of how Waymo builds trust with riders
(15:55) The commercial, operational, and system behavior metrics Waymo uses
(20:38) What are L5 autonomous vehicles and why Shweta thinks L4 vehicles are good enough
(22:53) How to keep investors enthusiastic when it’s a long-term investment
(25:24) Building successful teams and successful products
(26:39) Determining what you’re not building, especially before product-market-fit
(27:49) Why large companies need to disrupt their own models
(29:33) The most underrated product management skills
(33:07) Tips for getting promoted
(35:19) Where is Waymo and how to try it out
(36:46) Lightning round

 

[Doggydogworld]

So it still proves that Waymo can do L4 highways.

Waymo has tested on highways pretty much continuously since 2009. Last year Joel said he saw an uptick in Phoenix area highway testing. I don't consider it L4 until they pull the safety driver. Until then it's "L4 design intent". Same as Tesla (though they lie about that to the DMV).

I was just hoping I'd missed something and they'd actually pulled safety drivers on highways. But I should have figured otherwise from the video. They always make a point to show the driver's seat when it's empty.

What about Texas Toll Road. SH 130

Only one section of 130 is 85 mph -- the section that goes from nowhere to nowhere. They made it 85 mph to attract joyriders so they could at least claim some revenue. I used to drive the useful part of 130, which circumvents Austin traffic, every month or so back in 2014-2015 and still drive it once in a while. It goes right past Tesla's factory. Even the useful part didn't have many cars back in the day and I never saw a single vehicle coming on or off the 85 mph section. The useful part is 80 mph, btw, as are a bunch of other highways in TX. Especially west TX.

 

[ThomasD]

When I was Stationed at Cannon AFB we would drive to Amarillo a lot. Driving from Clovis to Amarillo we would average 80 any faster and the State Patrol would get you.

 

[diplomat33]

Waymo has tested on highways pretty much continuously since 2009. Last year Joel said he saw an uptick in Phoenix area highway testing. I don't consider it L4 until they pull the safety driver. Until then it's "L4 design intent". Same as Tesla (though they lie about that to the DMV).

You might not consider it L4 unless it is driverless but it is L4. The SAE does not make any distinction between L4 and L4 design intent. A system cannot be "L4 design intent" but not "L4". A system is either L4 or not. And if a system is L4, then it is always L4, whether there is a safety driver or not. Having a safety driver does not downgrade the L4 system to L2. The SAE is very clear about that. Tesla FSD is not "L4 design intent", Tesla FSD is simply L2.

Now we can make a distinction between testing and deployment. I think it is fair to say that a L4 system is still testing and not truly deployed to the public unless it is driverless. But that does not make it less L4. And we need to specify the ODD. Waymo has removed the safety driver in certain ODD but has a safety driver in other ODD. So we can say that Waymo is testing L4 in certain ODD but has deployed L4 to the public in other ODD.

I was just hoping I'd missed something and they'd actually pulled safety drivers on highways. But I should have figured otherwise from the video. They always make a point to show the driver's seat when it's empty.

I do think they will pull the safety driver from highway driving later this year. Just my guess though.

 

[diplomat33]

Well, it looks like Waymo had a 12 car stall event in Phoenix:

https://twitter.com/x/status/1645472114055729153

Obviously, these stall events are never good. It is a reminder that autonomous driving is not perfect. Hoping we get some answers from Waymo soon.

But we also need to put them in their proper context. These stall events are pretty rare compared to the total driverless miles that Waymo and Cruise are doing. And we need to distinguish between stalls that cause some traffic delays versus actual accidents. I agree with Brad Templeton, Waymo and Cruise should address these stalls and I am sure they are working on it but a few stalls per million miles is a small price to pay if we can ultimately save 40,000 lives per year. Transit Study Reveals Robotaxis Causing Surprisingly Little Disruption On Streets

 

[enemji]

That has nothing to do with capabilities and technology. A Tesla in San Francisco, LA, Pheonix is not as capable as a Waymo in those same areas regardless of if you take geofencing and obvious segmentation of the market into consideration.

Scaling is considerable easier than the creating the technologies required to drive and do so at an order of magnitude safety levels compared to human drivers. Creating new technologies and testing is the hard part. It has taken Waymo over 10 years and 4 generations of hardware and countless software iterations to be able to safely drive in the rain without a safety personnel in the car. It has taken Tesla about 10 years, several iterations of software and 4 generations of compute platforms and they still can't remove a driver from the driver seat, that should tell you where the hard problem is.

No, we can clearly see who actually has autonomous vehicles driving around. We can't say the jury is out on who is ahead. Same way we can't say the Jury is out on who is ahead in landing rockets. SpaceX is clearly ahead even though BlueOrigin and a few smaller rocket companies can do it, even NASA did it in the early 90s.

That is a strawman. Having L5 is not a prerequisite to see who is ahead. Waymo set out to create a L4 vehicle not L5. L5 is a goal to strive for but would not be achieved in the next 20 years.

Waymo is making a L4 vehicle that drives by itself. Tesla does not have any such vehicles doing that anywhere on this planet. Tesla is one of the many car makers selling vehicles with L2 ADAS.

Special Purpose vs General Purpose.

 

[diplomat33]

Just stop there. The difference between L5 & L4 is geo-scaling.

Not quite. The difference between L4 and L5 is ODD. But geo-scaling is not the only ODD limitation. So for example, you could have L4 that is available everywhere but is limited to good weather. So you can have geo-scaling and still have L4. L5 is more than just removing all geofences, it also needs to be safe and reliable in all weather than humans can handle, all traffic conditions that humans can handle, etc...

But personally, I don't think we even need L5. We could have L4 with a big enough ODD and that could be "good enough" for most people. For example, if we had L4 that works on 95% of US roads, that would likely be good for most people in the US. And certainly, L4 that works on 95% of US roads could be very profitable. So companies don't need L5 to be profitable. Why even do L5 if L4 in a big enough ODD is good enough? Certainly, we will have L4 in a big ODD way before we ever get L5.

 

[johnm]

Not quite. The difference between L4 and L5 is ODD. But geo-scaling is not the only ODD limitation. So for example, you could have L4 that is available everywhere but is limited to good weather. So you can have geo-scaling and still have L4. L5 is more than just removing all geofences, it also needs to be safe and reliable in all weather than humans can handle, all traffic conditions that humans can handle, etc...

But personally, I don't think we even need L5. We could have L4 with a big enough ODD and that could be "good enough" for most people. For example, if we had L4 that works on 95% of US roads, that would likely be good for most people in the US. And certainly, L4 that works on 95% of US roads could be very profitable. So companies don't need L5 to be profitable. Why even do L5 if L4 in a big enough ODD is good enough? Certainly, we will have L4 in a big ODD way before we ever get L5.

Do you have something specific in mind that would cause 5% of US roads to excluded? I am unable to come up with anything that seems reasonable.

 

[diplomat33]

Do you have something specific in mind that would cause 5% of US roads to excluded? I am unable to come up with anything that seems reasonable.

Well, the "5%" was really just to make a point that L4 that covered "almost L5" ODD would still be very useful and good enough. It does not have to be 5%.

But a company could release L4 that works everywhere except rural roads. Also, Mobileye crowdsources the fleet to build and update their maps for autonomous driving. So they have maps for everywhere that one of the consumer cars with their tech has driven. It's millions of cars so their maps cover a lot of roads. Mobileye has mapped almost every road in the US and EU now. But conceivably there could be some roads that none of their cars has ever driven and therefore has not been mapped yet. Even with millions of consumer cars, I doubt they cover every single road in the US. There could be some rural roads that none of the Mobileye powered cars have driven. That could cover ~5%.

And GM has said that their L2 "hands off" system called Ultra Cruise will cover 95% of driving scenarios.

General Motors announced today that Ultra Cruise, the company’s next-generation advanced driver assistance system2 designed to ultimately enable hands-free driving in 95 percent of all driving scenarios,

Site Maintenance

 

[EVNow]

Why even do L5 if L4 in a big enough ODD is good enough? Certainly, we will have L4 in a big ODD way before we ever get L5.

We have discussed this ad naseum. Going straight on a 1/2 mile strip of road with no intersections and traffic in good weather is also L4.

That is why levels are stupid.

 

[ThomasD]

Did they figure out what happened? Did the cars fix themselves or were they human driven out of the situation?

 

[diplomat33]

Did they figure out what happened?

We don't know. Waymo has not said what happened yet.

Did the cars fix themselves or were they human driven out of the situation?

The tweet says that Waymo techs came so I assume that means they were human driven out of the situation.