Waymo

[mikes_fsd]

Waymo does not have to fail for Tesla to succeed.

Agreed! Hence don't prescribe Lidar and other appendages to Tesla.

Prove that your approach works, Waymo! Don't tell others what THEY must do, re equipment and approach.

So far, they're proving that their approach is not going to work in the real world.
So, our resident Waymo apologist has to grab at straws to keep the topic afloat.

 

[diplomat33]

Agreed! Hence don't prescribe Lidar and other appendages to Tesla.

Prove that your approach works, Waymo! Don't tell others what THEY must do, re equipment and approach.

So far, they're proving that their approach is not going to work in the real world.
So, our resident Waymo apologist has to grab at straws to keep the topic afloat.

HA HA. Waymo has fully autonomous ride-hailing available to the public, millions of autonomous miles and L4 being tested in 25+ cities. But they need to prove to a company that only has L2 that their approach works? Waymo is proving that their technology works in the real world.

It is Tesla that needs to prove that their approach works. So far they still only have L2. They need to prove that their L2 approach can achieve L4 or L5. So Tesla, prove your approach works before you lecture others about using lidar.

 

[mikes_fsd]

HA HA. Waymo has fully autonomous ride-hailing available to the public

 

[mark95476]

The biggest tell for me was seeing all of those Waymos driving around Mountain View for years and not a single one was carrying Google/Waymo employees to the office and back. No dogfooding. It's been ~ a year since I've seen a Waymo in Mountain View, but they were always carrying two people, never any more or less.

 

[diplomat33]

The biggest tell for me was seeing all of those Waymos driving around Mountain View for years and not a single one was carrying Google/Waymo employees to the office and back. No dogfooding. It's been ~ a year since I've seen a Waymo in Mountain View, but they were always carrying two people, never any more or less.

That's how you judge FSD? LOL.

 

[Doggydogworld]

We are testing trucking between Phoenix and Tucson.

Lie! She said Tuscon to Texas.

Only in a subset of Chandler can get a van without a safety driver.

Lie! The service area includes parts of both Tempe and Mesa.

C'mon people. Real people just say Phoenix, not "Phoenix-Mesa-Scottsdale Metropolitan Statistical Area". The core issue is Waymo launched a ~50 square mile driverless Robotaxi operation in 2018 and they serve the same ~50 square mile area in 2021. They have not scaled. They have no immediate plans to scale. They aren't going to tell us why they can't scale, no matter how many times the Bloomberg cutie or people on Reddit ask. And they aren't going to tell us when they might scale, either, assuming they even know.

We are left to guess. Some guesses are funny ("mapping is too hard", lol), others are serious but unsupported. Krafcik last year said safety wasn't the issue:

“And how do we know we’re ready? We’ve had our wonderful group of early riders, who’ve helped us hone the service, obviously not from a safety standpoint because we’ve had the confidence on the safety side for some time, but rather more for the fit of the product itself.”

Even if you don't believe Krafcik, Waymo is ultra-safety conscious and wouldn't operate driverless at all if it was unsafe. And if it's safe in those 50 square miles then it's safe in millions of other similar square miles with good weather and wide, well-marked roads. The same is true for regulation, which eliminates some locales but leaves plenty of others wide open.

It's obivously a business model problem. Revenue is far below expenses, and they don't know how to fix it. Entrepreneurs experiment wildly when this happens, until they stumble onto something that works (or they go broke). Waymo has no entrepreneurs, though, and no funding cliff which forces them to frantically experiment. So they just keep heading in a straight line. Their cars don't "ride on rails", but their business thinking does.

 

[diplomat33]

It's obivously a business model problem. Revenue is far below expenses, and they don't know how to fix it. Entrepreneurs experiment wildly when this happens, until they stumble onto something that works (or they go broke). Waymo has no entrepreneurs, though, and no funding cliff which forces them to frantically experiment. So they just keep heading in a straight line. Their cars don't "ride on rails", but their business thinking does.

Doing nothing and just continuing to operate the same ride-hailing in Chandler at a loss, month after month, is not an option IMO.

Personally, I think Waymo should just "go for it" and expand to another city. They should just launch another ride-hailing service in a 50-100 sq mi area in say San Francisco or Las Vegas with maybe a few hundred I-Pace's to start off. They could start with safety drivers for say the first 6 months just to make sure everything goes well. They could limit the ODD to good weather too if they are worried about that. Then, as the safety is proven, they could gradually remove the safety drivers. At least this way, they would show movement in the right direction. They could at least say that they've expanded Waymo One to a new area. And Krafcik has claimed that the 5th gen is cheaper to scale. If that's true then a ride-hailing service with all i-Pace's should be cheaper to operate and make it possible to start making a profit.

 

[diplomat33]

The AMA is live.

How are you going to scale? What are the impediments to service expansion at this time?

I’ll start with the most important point - we are going to scale safely. Within Waymo, we have a phrase that “safety is urgent.” We feel the same urgency to scale quickly that others do, but a ton of work goes into doing it safely...including building public trust, working with regulators and community stakeholders, and scaling what we believe will be a very successful service at the right time.

What I’ve found really impressive as we’ve tested in cities across the US is that each new situation we encounter in one part of the world helps us handle similar situations elsewhere. So the gains we’ve made compound and therefore reduce the time it takes to drive in a new environment. We take the things we’ve learned in one location and apply them to the next. And this is how we’re going to scale - build up experiences so we can scale faster (but still safely) as we bring our technology to more places.

In terms of impediments, I wouldn’t really call them that. We have a really clear roadmap on what it takes to safely bring a fully autonomous service to market. We start by sending out our mapping team to collect local data and build high quality maps. Then we start testing. And testing and testing! and testing. Including the edge cases. It’s all about rounds and rounds of validation to help ensure the Waymo Driver is ready for fully autonomous operation. Even developing these frameworks and methodologies takes tremendous rigor. We test on public roads with an autonomous specialist (i.e. safety driver) on hand, we augment this with simulated driving to test under a variety of new and novel conditions, as well as structured testing at our test tracks.

Once the Waymo Driver is ready, we begin testing and validation without the autonomous specialist. And while all this is happening, we work on the nuances of the rider-only experience - pickups and dropoffs, app experience, features, etc. (Sam).

Waymo has said before that once 5th gen vehicles have ramped up, expansion will be swift. What specifically does the 5th gen platform offer that makes it ideal to scale?

Our 5th gen hardware has amazing capabilities. Its enhanced sensor suite is able to provide our software with even more data so our Waymo Driver can evaluate increasingly complex situations. It’s also designed to tackle more diverse conditions, such as additional types of weather. And because it’s built to work on different vehicle platforms, it brings development and cost efficiencies which are critical to scaling. (Sam)

I just want to know when you're going to start driving to the ASU campus. That would make my commute a lot easier!

We are testing there right now so you may see us around. No exact date….but I’ll let the team know you asked for it! (Lety)

 

[ThomasD]

By the time Waymo maps every city and town in the country my Grandkids will have Grandkids

 

[DanCar]

Lie! The service area includes parts of both Tempe and Mesa.

You've succumbed to Waymo lies. Find a video where there is a passenger riding with no safety driver in Tempe or Mesa. They are all in a subset of Chandler.

 

[diplomat33]

By the time Waymo maps every city and town in the country my Grandkids will have Grandkids

It does not take that long to map areas. It only takes a few days to map an entire city. Also, Waymo will offer ride-hailing in major metro areas first. So they don't need to map the entire US right away. They can start with mapping the metro areas where they offer ride-hailing first.

 

[diplomat33]

More from the AMA.

What do you see as your next potential markets? Are they foreign or domestic?

Cities with heavy ride-hailing usage are our priority and right now we’re focused on the U.S. first. (Lety)

How well does Waymo currently handle rain and snow? Most videos of rides I have seen have led to a driver taking over, but I am curious whether that is a computer vision problem, or simply being overcautious as the vehicle itself is less safe in that weather?

Our 5th gen hardware was especially designed to handle rain and snow and we continue to test in these environments. Here’s some testing from the other day of our I-PACE (Lety)

How well does Waymo currently handle rain and snow? Most videos of rides I have seen have led to a driver taking over, but I am curious whether that is a computer vision problem, or simply being overcautious as the vehicle itself is less safe in that weather?

Yea weather creates new challenges along both those dimensions, for humans as well as robots. For example, it's harder to accurately identify objects on the road during snow accumulation, and slippery roads create new challenges for motion control. We also have to think about things like how we clean the sensors. We're doing targeted testing on all kinds of weather across the US (e.g. Kirkland, Detroit, Miami) so that the Waymo Driver is well suited to handle all kinds of environments as we scale (Sam)

What would you say Waymo's position is in regards to competition from Mobileye, Cruise, Tesla, Autox, Zoox, and other auto manufacturers? What do you see as your competitive advantage?

Lots of companies out there are doing exciting work, with everyone taking their own spin on their core tech and the products that they're building.

But Waymo has a huge advantage in terms of experience developing fully autonomous technology (aka no one in the driver’s seat). It takes a LOT to go from having a backup driver to having an empty driver’s seat...safely. You don’t know what you don’t know, until you’ve been there like we have. (Sam)

How do you balance expanding too early vs too late? What possible events in the future would cause you to think you went too fast, in hindsight? What possible events in the future would cause you to think you went too slowly, in hindsight?

It’s exactly that - a balance. I mentioned earlier how we’re always reminding ourselves at Waymo that safety is urgent - we want to make roads safer by deploying our technology, and the sooner we do the quicker we can reduce the number of fatal crashes due to human error like distraction, impairment, and drowsiness. But we need to make sure we deploy safely as well. You can catch a sense of how we find this balance between going too fast vs. going too slow from the safety report and whitepapers on our website. (Sam)

 

[powertoold]

Holyyyy, what the hell did I just watch

 

[diplomat33]

Holyyyy, what the hell did I just watch

Waymo has a track facility where they do some testing. You watched a 7 second clip of the some of the track testing. The video is kinda useless to be honest since it is so short.

This clip is a bit more informative IMO as it shows the car maneuvering in a narrow street in SF.

 

[Doggydogworld]

You've succumbed to Waymo lies. Find a video where there is a passenger riding with no safety driver in Tempe or Mesa. They are all in a subset of Chandler.

LOL. This ride ends at Fiesta Crossing in Mesa. Early in the video he shows the app with the service area. Waymo may not have an entire cult stalking them, but I still think word would get out if they "lied" about their service area.

Better pics of the service area are at the bottom of the JJRicks FAQ page.

This AMA was better than the last one. Still no real answers to the main questions about expansion, of course, but a few interesting details.

 

[powertoold]

It does not take that long to map areas. It only takes a few days to map an entire city.

I'm not going into detail since I'm tired of these discussions, but I think you underestimate the amount of information (and also validation) recorded / needed in HD maps. It's not only about "mapping" like generating a 3D map of the environment. You also need people to label and verify lane semantics / parking areas vs lanes / speed limit signs / no turn on red signs / arrows / right only lanes / left only lanes / etc. etc. All of this needs to be validated as well, by having safety drivers drive the HD map to verify correct labels.

I'll give you one example:

Your mapping car drives around to map the city, but on one part of the road, a large truck is next to you and blocking a "right turn only" lane sign / marking...

 

[DanCar]

LOL. This ride ends at Fiesta Crossing in Mesa. Early in the video he shows the app with the service area. Waymo may not have an entire cult stalking them, but I still think word would get out if they "lied" about their service area.

Better pics of the service area are at the bottom of the JJRicks FAQ page.

This AMA was better than the last one. Still no real answers to the main questions about expansion, of course, but a few interesting details.

My bad. I see it does include a tiny sliver of Mesa. Thanks for the corrections.

 

[diplomat33]

I'm not going into detail since I'm tired of these discussions, but I think you underestimate the amount of information (and also validation) recorded / needed in HD maps. It's not only about "mapping" like generating a 3D map of the environment. You also need people to label and verify lane semantics / parking areas vs lanes / speed limit signs / no turn on red signs / arrows / right only lanes / left only lanes / etc. etc. All of this needs to be validated as well, by having safety drivers drive the HD map to verify correct labels.

I'll give you one example:

Your mapping car drives around to map the city, but on one part of the road, a large truck is next to you and blocking a "right turn only" lane sign / marking...

Obviously, my "few days" is an educated guess. It would depend on several factors like the size of the mapped area, how many cars you have mapping, your computing for mapping etc...

I know there is a lot of additional information that goes into the maps. But labelling is done automatically now. Most, if not all, of the HD mapping process is automated now.

NavInfo has automated HD mapping:

Yes, you would probably drive around some more after the HD mapping is done, with safety drivers, to validate the map.

In your case of the right turn sign being blocked a truck, you could probably just do another pass when the truck is gone. Multiple cars drive the same route for this very reason. Or even just add the sign on the map manually later.

 

[powertoold]

Yes, you would probably drive around some more after the HD mapping is done, with safety drivers, to validate the map.

In your case of the right turn sign being blocked a truck, you could probably just do another pass when the truck is gone. Multiple cars drive the same route for this very reason. Or even just add the sign on the map manually later.

After I thought about HD maps for a while, it seems they have to be way more detailed than my first impression.

For example, there's a speed limit, but there may also be "common sense" speeds that you need to label on the HD map:

In SF, the limit may be 25mph, but you really don't want to drive 25mph into the beginning of a steep hill.

On a curve with parked cars on the right, you may want to slow down in case there are peds or people opening their car doors to get out. ALSO your speed may be different if you drive on the inside lane vs outside lane on some curves.

There's a lot of these types of situations that need to be labeled on an HD map. AND it may be nuanced in every city / state / locale.

 

[Bladerskb]

After I thought about HD maps for a while, it seems they have to be way more detailed than my first impression.

For example, there's a speed limit, but there may also be "common sense" speeds that you need to label on the HD map:

In SF, the limit may be 25mph, but you really don't want to drive 25mph into the beginning of a steep hill.

On a curve with parked cars on the right, you may want to slow down in case there are peds or people opening their car doors to get out. ALSO your speed may be different if you drive on the inside lane vs outside lane on some curves.

There's a lot of these types of situations that need to be labeled on an HD map. AND it may be nuanced in every city / state / locale.

You mean you got that from Amon Shashua's presentation?
lol HD map automation and HD map at scale has already been solved using deep learning and crowd sourcing.
Anyone who still says otherwise is spreading FUD. Mobileye's REM is 100% fully automated all over the entire world.