Waymo

[Bladerskb]

You're assuming the occupancy network is trained like 2D segmentation, when we know that's not the case. It's trained on multiple, overlapping views with a defined time dimension. If what the network is ultimately detecting is 3D volume, it doesn't need to have a similar representation of an object in the training data to know that a 3D volume is perceived in a particular way via bi/trinocular views over multiple successive frames.

E.g. a garbage bag blowing across the street will be perceived by the occupancy network as a 3D volume because of the parallax in the multiple camera views telling it that those pixels are closer to the vehicle than the background, and how that parallax changes as the bag itself moves and the vehicle camera views move toward it.

I didn't say representation of the object. I said 'similar representation of what it sees'. I will explain later (work)

Got this Tweet as a promotion from Mobileye today, I thought it was a really interesting point of comparsion:

https://twitter.com/x/status/1559139372196323328

You can really see the limitations of traditional image-space segmentation, as highlighted by Ashok's CVPR presentation. They seem to be doing the vehicle and driveable space segmentation in individual 2D views.

You can clearly see:
1. The density at the horizon causing a lot of noise in the distance of the VIDAR representation
2. The VIDAR representation exhibiting some fisheye warping, presumably due to the stitching together of of individual 2D viewpoints
3. Driveable space not persisting behind vehicles and other obstacles

In addition to what i said earlier. In context of this current topic, Its worth pointing out that this is nowhere close to the resolution and recall you get with modern lidar.

&t=11601s

 

[diplomat33]

JJ Ricks spied some Waymo I-Pace cars testing on the highways in the Phoenix area and east of the current area. My guess is Waymo is getting ready to connect the downtown and Chandler areas and expand the service areas too, including the highways. I do think it is encouraging for scaling because Waymo can do suburban driverless and urban driverless. So once they can do highway driverless too, they should be able to expand to a lot more areas.

https://twitter.com/x/status/1562851613064105985

https://twitter.com/x/status/1562853652829679616

 

[diplomat33]

For reference, here is a map of the current Waymo service areas in Phoenix, marked in black. I marked on the map, red arrows to indicate the approx. location and direction of travel of the Waymos that JJ saw.

I know it is speculation but it seems to me like Waymo might be wanting to add highway 60 and 10 to connect downtown and Chandler areas. We know they want to add Sky Harbor. The other arrow could indicate that they also want to add the 202 and expand east past the Chandler area, maybe add Gilbert. I wonder if they want to add the Phoenix Mesa Gateway airport.

 

[Supcom]

If Waymo could expand to cover five entire states per year, it would still take ten years to provide coverage to everyone.

Sadly, Waymo doesn't even cover one entire city. They kind of need to step up the pace a little bit.

 

[diplomat33]

If Waymo could expand to cover five entire states per year, it would still take ten years to provide coverage to everyone.

Sadly, Waymo doesn't even cover one entire city. They kind of need to step up the pace a little bit.

I think it is a mistake to assume expansion will always continue at the same pace we've seen so far. The first cities are going to be slower just because Waymo has never done it before but as Waymo expands to new areas, expansion should go faster.

One, Waymo is adding more and more skills and reliability to the Waymo Driver. As the Waymo Driver becomes more competent and more experienced, it will be able to drive reliably in more areas. Those driving skills will transfer to other places. For example, the driverless in Chandler will apply to other similar suburban areas, the driverless in SF will apply to other similar urban areas, etc... Waymo seems to be taking a strategic approach to scaling. They focused on light urban driving (Chandler), medium urban driving (San Franscisco), now they are testing in heavy urban driving (NYC) and testing highway driving (Phoenix & Texas). They have good driverless in the first two domains now. Once they have reliable driverless in all four domains, they should be able to scale to any city in the US.

Second, Waymo will learn what worked or did not work with testing, validation and deployment and will get more efficient at the deployment process. So in theory, Waymo should be able to scale faster with each new area.

How quickly Waymo is able to turn theory into practice, we shall see. Personally, I do think we will see expansion go much faster soon. I would also point out that while driverless everywhere is the ultimate goal, Waymo can be profitable and their tech can provide real benefits way before they cover every part of the US.

 

[diplomat33]

Another driverless video from a Waymo employee. This one is edited but real-time. It does show a couple interesting cases IMO.

 

[ThomasD]

Is there going to be a way to move the front seats forward. Giving more room to taller/larger people. With the front seats that far back I don't think I could get into the backseat

 

[diplomat33]

Is there going to be a way to move the front seats forward. Giving more room to taller/larger people. With the front seats that far back I don't think I could get into the backseat

Yes, of course.

 

[Dewg]

Was having a discussion in another thread and a scenario came up about a traffic intersection being controlled by a traffic cop. How would Waymo handle an intersection where the lights are out and a cop is waving people?

 

[Daniel in SD]

Was having a discussion in another thread and a scenario came up about a traffic intersection being controlled by a traffic cop. How would Waymo handle an intersection where the lights are out and a cop is waving people?

They claim to be able to handle it:

I suspect that when they're operating without a safety driver they have remote assistance confirm the cars actions.

 

[ThomasD]

Are they able to recognize a School bus stopped on a multi lane highway ? Where the School Bus is in the far right lane and the Waymo is in the left lane

 

[Daniel in SD]

Are they able to recognize a School bus stopped on a multi lane highway ? Where the School Bus is in the far right lane and the Waymo is in the left lane

You can ask them on Twitter. They do seem to reply.

 

[diplomat33]

Are they able to recognize a School bus stopped on a multi lane highway ? Where the School Bus is in the far right lane and the Waymo is in the left lane

yes, they can handle that scenario. They can definitely recognize a stopped school bus. And the planner would know what to do. That's a simple scenario.

 

[diplomat33]

They claim to be able to handle it:

I suspect that when they're operating without a safety driver they have remote assistance confirm the cars actions.

That video is from 2019. I think by now on the 5th gen, the car would be able to handle it without needing remote assistance.

 

[SO16]

HA HA. Very funny.

The operating areas are not stamp sized. They cover Chandler and downtown Phoenix as well as a good chunk of SF. And since mapping is cheapy and easy, there is no reason the operating area can't be much larger, even the entire US. The only limit to Waymo scaling is when they decide the safety is good enough in the operating area. There are lots of different edge cases in different US cities.

If you took all the areas and roads where driverless cars will be intended to operate and shrunk that down to a post card, the Waymo mapped area would probably be equivalent to the size taken up by the copyright logo in fine print on the post card. The postage stamp size is being generous.

 

[diplomat33]

If you took all the areas and roads where driverless cars will be intended to operate and shrunk that down to a post card, the Waymo mapped area would probably be equivalent to the size taken up by the copyright logo in fine print on the post card. The postage stamp size is being generous.

For now maybe. But Waymo is testing in way more areas though. The driverless areas will expand over time. Eventually driverless cars will operate everywhere. Remember driverless cars need to be able to do 1M hours of driving per safety critical mistake. That's a very high bar. So it makes sense that driverless areas are very small right now. But those driverless areas will get bigger as AV companies solve more autonomous driving and deploy more driverless cars.

 

[SO16]

For now maybe. But Waymo is testing in way more areas though. The driverless areas will expand over time. Eventually driverless cars will operate everywhere. Remember driverless cars need to be able to do 1M hours of driving per safety critical mistake. That's a very high bar. So it makes sense that driverless areas are very small right now. But those driverless areas will get bigger as AV companies solve more autonomous driving and deploy more driverless cars.

I’m curious, how much does Waymo pay you and how can I get in on that gig?

 

[diplomat33]

I’m curious, how much does Waymo pay you and how can I get in on that gig?

I work pro bono.

 

[SO16]

I work pro bono.

They are lucky to get your services for free. You certainly put in a boatload of effort for them.

 

[diplomat33]

They are lucky to get your services for free. You certainly put in a boatload of effort for them.

I am just a fan of autonomous driving.