Waymo

[diplomat33]

Dolgov just tweeted that Waymo uses the same 5th Gen cars that drive autonomously to also map an area. He also posts a little clip of autonomous driving in LA to show what the lidar sees:

https://twitter.com/x/status/1443350735479971843

The other tweet suggests that Waymo is actively testing autonomous driving in LA.

 

[diplomat33]

Since Waymo uses the same 5th Gen cars for ride-hailing to also map an area, it got me thinking. Waymo could sell a 5th Gen car to consumers and let the consumer automatically map the areas where they like to drive. They would simply drive around an area for a little bit, similar to AP camera calibration period when you first activate AP, and then after the mapping is done, the car would allow you to activate the Waymo Driver. I'd definitely be ok with an autonomous car that maybe needs me to drive around manually in the beginning, if I am in a new area that has not been mapped yet, and then afterwards, let's me go L4. For example, I could map my daily commute first and then enjoy L4 on my daily commute. And the maps would automatically grow as the fleet grows.

Obviously, cost is still an issue. The cost would need to come down. Also, Waymo does not want consumers to supervise the system, so the FSD would need to be good enough for no supervision before they could sell a consumer car. So Waymo would need to improve the FSD more before they could do this. But at least, mapping is not an obstacle.

 

[Doggydogworld]

Co-CEO Tekedra Mawakana interview at Code Conference. Transcript so you don't have to spend 36 minutes listening.

Eventually we will also focus on personal car ownership, which is when you license the technology.

Confirmation that they still plan to license their system for personal cars, as Krafcik said for years. Some here have repeatedly claimed otherwise. I recommend ignoring anything they say about Waymo.

She still claims 20 million miles, which is way out of date. This is intentional. They updated the number regularly when it was "increasing exponentially". But linear growth tells a story they don't want you to hear. And they definitely don't want analysts and investors to extrapolate linearly looking forward.

Our goal in Phoenix was to learn.

Sorry, but no. Krafcik made it clear, and the 82k vehicles they "ordered", the factory they built, etc. confirmed that the original goal was a full commercial launch in greater Phoenix, followed by other cities. It only turned into a research project when customers failed to materialize.

The technology was ready, as proven when they pulled safety drivers. It's not like the rest of Phoenix presents unique showstopper challenges they don't already face in Chandler. But you can't spend billions on vehicles and support infrastructure unless you have customers. Lots of customers. That's the missing ingredient. And that's why they're in San Francisco now. It's more technically challenging, but at least there's proven demand.

 

[diplomat33]

Confirmation that they still plan to license their system for personal cars, as Krafcik said for years. Some here have repeatedly claimed otherwise. I recommend ignoring anything they say about Waymo.

Thanks for the transcript.

Yes, Waymo is interested in personal car ownership. Ride-hailing allows them to deploy something for the public now while they continue to improve the technology. But they will do personal car ownership at some point. Personally, I think they are waiting for the FSD to be better and for the cost to come down more. Waymo does not want the owner to have to supervise the system, so they need the L4 to be better where they can completely remove driver supervision first. And the cost has to come down to make it more affordable. Once they achieve these two goals, they will do personal car ownership, I have no doubt.

Sorry, but no. Krafcik made it clear, and the 82k vehicles they "ordered", the factory they built, etc. confirmed that the original goal was a full commercial launch in greater Phoenix, followed by other cities. It only turned into a research project when customers failed to materialize.

The technology was ready, as proven when they pulled safety drivers. It's not like the rest of Phoenix presents unique showstopper challenges they don't already face in Chandler. But you can't spend billions on vehicles and support infrastructure unless you have customers. Lots of customers. That's the missing ingredient. And that's why they're in San Francisco now. It's more technically challenging, but at least there's proven demand.

I think you are both right. But Waymo did learn a lot from Phoenix about deploying and managing a ride-hailing service, customer service, etc... So there was definitely a learning component in Phoenix as well.

 

[EVNow]

The technology was ready, as proven when they pulled safety drivers. It's not like the rest of Phoenix presents unique showstopper challenges they don't already face in Chandler. But you can't spend billions on vehicles and support infrastructure unless you have customers. Lots of customers. That's the missing ingredient. And that's why they're in San Francisco now. It's more technically challenging, but at least there's proven demand.

That is an interesting take.

But one thing I disagree on - they should have extended the area to include more complex parts of Pheonix. They never did that. Should have been very simple if they didn't have any technical challenges.

Eventually we will also focus on personal car ownership, which is when you license the technology.

To me thats not a "plan". Thats the eventual goal. Its like Musk's idea to colonize Mars. He doesn't have any "plans" for that, yet - just a dream / goal.

 

[diplomat33]

Tomorrow and Thursday will be big days for Waymo. Both Dolgov and Anguelov will be talking about Waymo's AI and ML. I expect we will get a lot of updates on the technical progress that Waymo has made.

In this keynote, Drago Anguelov, Head of Research at Waymo, discusses Waymo’s progress towards building a scalable technology stack for autonomous driving vehicles. With more than a decade of experience in solving autonomous driving, Waymo is now operating the world’s first commercial ride-hailing service Waymo One in Phoenix and has recently welcomed its first riders in San Francisco by kicking off the Trusted Tester program.

Drago will give an overview of the key autonomous driving challenges and describe how Waymo is leveraging the cutting edge ML systems across the stack to handle them. He will also outline promising avenues to keep expanding the scope of ML in the stack in the future and showcase some of Waymo’s work in the space.

https://twitter.com/x/status/1445063742421716996

https://twitter.com/x/status/1445426059147677708

 

[powertoold]

"It's going great!... " - Tekedra

As she was talking about the Google self-driving project in 2013 and how employees didn't pay attention, I had an obvious epiphany. Tesla is in a different paradigm when it comes to driver attention. Tesla drivers own their cars. They're fully aware of their liability.

Google self-driving cars were owned by Google, and the employees knew that most liability fell on Google.

 

[diplomat33]

Tesla is in a different paradigm when it comes to driver attention. Tesla drivers own their cars. They're fully aware of their liability.

And yet we've see Tesla owners being very irresponsible with driver attention like this "sleeping on AP" stunt:

 

[powertoold]

And yet we've see Tesla owners being very irresponsible with driver attention like this "sleeping on AP" stunt:

Almost everything related to fsd development is a statistic. My point was that Tesla owners are much more likely to pay attention vs Google employees because they have higher risk to personal property and liability.

And that's why we see such great autopilot safety statistics coming out from Tesla. Apparently autopilot plus Tesla owner is much safer than human alone.

 

[diplomat33]

Almost everything related to fsd development is a statistic. My point was that Tesla owners are much more likely to pay attention vs Google employees because they have higher risk to personal property and liability.

And that's why we see such great autopilot safety statistics coming out from Tesla. Apparently autopilot plus Tesla owner is much safer than human alone.

Statistically, you might be right that Tesla owners are more likely to pay attention. However, there are lots of examples of Tesla owners not paying attention. I am sure if you captured the in-cabin camera footage from Teslas, you could find lots of examples of Tesla owners doing the exact same thing as those Google employees (texting, checking their phone, putting on make-up etc).

It should also be noted that Tesla requires hands-on wheel whereas the Google experiment allowed hands-off. So that is a big difference. If Tesla allowed hands-off, we might see more cases of Tesla owners not paying attention, just like the Google employees. I doubt if Tesla allowed us to go hands-off that we would all pay attention perfectly 100% of the time just because we are liable.

The bottom line is that Google concluded that hands-off L2 was not a good idea because drivers will get complacent and not pay attention as much as they should. As a result, they cancelled their L2 system and focused on L4 exclusively, arguing that it is better to develop a system, that when finished, won't require any driver supervision.

(edited for clarity and typos).

 

[diplomat33]

Here is the video of the fireside chat with Waymo co-CEO Dolgov today at TransformX Conference:

Data-Driven AI for Autonomous Vehicles With Dmitri Dolgov of Waymo - Video | Scale Events

Dmitri Dolgov is the co-CEO of Waymo, an autonomous driving technology company. He is also one of the founders of the Google self-driving car project, which began in 2009 and became Waymo in 2016. He joins Scale AI CEO Alexandr Wang in a fireside chat to

exchange.scale.com

Very informative!

 

[diplomat33]

So, from Chandler to Sanfrancisco they took 3 years (?). How long before they add the next city ? How long before they add all top 100 US cities ?

During the fireside chat yesterday, Dolgov mentioned that they started ramping up testing in SF "earlier this year". They announced the trusted rider program at the end of August. Assuming "earlier this year" means January, that gives us about 8 months between ramping up testing in a new city and launching an "early access" ride-hailing service. That does give us another data point to help us gauge their expansion rate.

Obviously, 8 months per city would still be too long. But they went from years to launch in Chandler to 8 months to launch in SF. That is a pretty big improvement IMO. Waymo is accelerating the rate at which they can expand to a new city. I suspect the 3rd, 4th, cities will take less than 8 months. My guess is that eventually Waymo will be able to add new cities every 1-2 months and then maybe faster than that.

 

[rxlawdude]

Statistically, you might be right that Tesla owners are more likely to pay attention. However, there are lots of examples of Tesla owners not paying attention. I am sure if you captured the in-cabin camera footage from Teslas, you could find lots of examples of Tesla owners doing the exact same thing as those Google employees (texting, checking their phone, putting on make-up etc).

It should also be noted that Tesla requires hands-on wheel whereas the Google experiment allowed hands-off. So that is a big difference. If Tesla allowed hands-off, we might see more cases of Tesla owners not paying attention, just like the Google employees. I doubt if Tesla allowed us to go hands-off that we would all pay attention perfectly 100% of the time just because we are liable.

The bottom line is that Google concluded that hands-off L2 was not a good idea because drivers will get complacent and not pay attention as much as they should. As a result, they cancelled their L2 system and focused on L4 exclusively, arguing that it is better to develop a system, that when finished, won't require any driver supervision.

(edited for clarity and typos).

You can't use people acting like jackasses to deny a product from coming to market.

See knives, guns, dumb cruise control, pressurized whipped cream, etc.

 

[ItsNotAboutTheMoney]

You can't use people acting like jackasses to deny a product from coming to market.

See knives, guns, dumb cruise control, pressurized whipped cream, etc.

... laundry detergent divided into small plastic blobs.

Google also had very little financial motivation to be interested in anything less than L4.

 

[EVNow]

tatistically, you might be right that Tesla owners are more likely to pay attention. However, there are lots of examples of Tesla owners not paying attention. I am sure if you captured the in-cabin camera footage from Teslas, you could find lots of examples of Tesla owners doing the exact same thing as those Google employees (texting, checking their phone, putting on make-up etc).

You mean kind of things people do driving normal cars ?

BTW, I do expect people with FSD beta to drive much more carefully than minimum wage Waymo employees (probably contractors ?). Most of us drive only 5 to 10 miles in the city per trip. Rest would be on freeways. Lot easier to concentrate for those 5 to 10 miles so that our expensive cars don't get into an accident.

Compare to a waymo minimum wage employee who has to spend hours and hours behind what is apparently a safe autonomous vehicle. If there is a crash - the employee will have to probably find a different minimum wage job. Not a big deal for many.

Definitely not the same.

 

[FloridaJohn]

You mean kind of things people do driving normal cars ?

BTW, I do expect people with FSD beta to drive much more carefully than minimum wage Waymo employees (probably contractors ?). Most of us drive only 5 to 10 miles in the city per trip. Rest would be on freeways. Lot easier to concentrate for those 5 to 10 miles so that our expensive cars don't get into an accident.

Compare to a waymo minimum wage employee who has to spend hours and hours behind what is apparently a safe autonomous vehicle. If there is a crash - the employee will have to probably find a different minimum wage job. Not a big deal for many.

Definitely not the same.

Waymo safety drivers get paid around $24/hour according to Glassdoor. They also go through intensive training for their jobs.

How much is Tesla paying it's beta testers? What type of training do we have to go through once we get approved for the Beta?

 

[diplomat33]

Here is Anguelov's keynote presentation:

ML at Waymo: Building a Scalable Autonomous Driving Stack with Drago Anguelov - Video | Scale Events

In this keynote, Drago Anguelov, Head of Research at Waymo, discusses Waymo’s progress towards building a scalable technology stack for autonomous driving vehicles. With more than a decade of experience in solving autonomous driving, Waymo is now operatin

exchange.scale.com

He shows some new clips of the 5th Gen autonomous driving in SF, explains how Waymo's FSD thinks, explains use of high res cameras, lidar and radar for best perception, rich and dynamic prior HD maps that can be shared in near real-time with the fleet, shows the high resolution of the 5th gen lidar, 3D point clouds with camera vision, ML for behavior prediction and planning, data augmentation, semantic point generation, self-supervised learning, imtation learning, high capacity ML models, and Waymo's simulation and how they are developing realistic AI agents to make simulation more realistic. He also shows an edge case where a high speed motorcycle that had lost it's driver, sped through an intersection and hit a trailer.

Looking at the clips of the 5th Gen, it seems to handle situations really smoothly, navigating very narrow streets with parked cars and pedestrians and cyclists cutting in. It is also interesting to see the little icons above objects that shows the car understands additional info about objects. The 5th Gen also understands gesturing of cyclists that they want to make a turn. IMO, the 5th Gen is much better than the 4th Gen in Chandler. We also see an example of night driving where camera vision may be a little hard to see but lidar is perfect.

 

[Daniel in SD]

You mean kind of things people do driving normal cars ?

BTW, I do expect people with FSD beta to drive much more carefully than minimum wage Waymo employees (probably contractors ?). Most of us drive only 5 to 10 miles in the city per trip. Rest would be on freeways. Lot easier to concentrate for those 5 to 10 miles so that our expensive cars don't get into an accident.

Compare to a waymo minimum wage employee who has to spend hours and hours behind what is apparently a safe autonomous vehicle. If there is a crash - the employee will have to probably find a different minimum wage job. Not a big deal for many.

Definitely not the same.

To be fair to Waymo safety drivers 14 of the 18 collisions in 6.1 million miles were getting rear ended. They got backed into once, cyclist ran into them once, sideswiped once, and one accident described as "straight crossing paths" (that one had an airbag deployment). Overall it looks like they get rear ended way more than the average driver but it's still an impressive performance. They avoided 29 collisions.
They only managed to avoid 2 rear end collisions. I'm guessing they probably don't usually intervene if the car seems to stop for no reason.

 

[EVNow]

To be fair to Waymo safety drivers 14 of the 18 collisions in 6.1 million miles were getting rear ended.

So, FSD with safety driver is very safe. Unlike what Waymo wants us to believe.

 

[Daniel in SD]

So, FSD with safety driver is very safe. Unlike what Waymo wants us to believe.

Maybe, depending on the system, the safety drivers, and the system for monitoring the safety drivers. 6.1 million miles is nowhere near enough to say though.