Waymo

[Bladerskb]

Service area, compared to old Uber/Lyft ride frequency map:
View attachment 711975

This is misleading as Waymo covers ~80% of SF. If you watch Tesla's FSD SF videos you would know that most of the areas in those videos where Tesla FSD fails spectacularly is covered in the Waymo's map coverage.

These are areas that Tesla's FSD can't go more than acouple miles without a safety disengagement. Those are the areas that Waymo covers.
The comments and commentaries should be. Waymo is about to go driverless in roads that Tesla can't even complete a single drive without trying to crash.

The real question people should be asking is. If Tesla is so good and "phoenix" is so easy and Tesla can do "chandler like performance" why does tesla FSD struggle to complete a drive in Waymos's so called "pathetic zone" according to Tesla fans?

Anyway, for any reasonable person. Waymo's coverage extends to 80% of SF. Which is great. Here is the boundary of SF.
You can click on the link so you can zoom in. For a good reference SF is 47.355 sq mile. This shows you how big Waymo's geofence in chandler was (50 sq mile).

Zoom in to street level to realize how big this geo-fence is!

Some people dont even know what 1 sq mile looks like. A big time Tesla fan recently compared Waymo's geofence in phoenix to the a-couple streets course in the Darpa Urban Challenge.

Here are the acouple streets of the Urban Grand Challenge with a mini highway. The yellow lines in the center.

Here is what 1 sq miles looks like:

Also that heat map you posted is from 2016 and just contains acouple of weeks. However look at what it looks like on Saturday:
Then when you look at the legend you realize this narrative is nonsense. Waymo won't even have not any close to enough cars to supply the demand in its ~80% coverage of the city.

 

[Bladerskb]

" Just look at what's going on in Chandler (nothing)."

That's not chandler, that's in California which makes sense as Gen 5 replaces Gen 4.
The same will happen in Chandler not too long from now.

 

[Bladerskb]

Seems plausible. How much do you estimate that would cost per mile?
What's so dumb about these endless Waymo vs. Tesla arguments is you're comparing an actual (barely) working system to a system that does not exist yet. My take away is that the reason for Waymo's limitations is that it's an incredibly difficult engineering problem and removing limitations wouldn't make it any easier to solve.

Most of this is automated with machine learning this goes for Waymo and of course for Mobileye as REM mapping is fully automated using ML, both the gathering of the map data, the building of the map and the publishing.

1 hr 47 mins 18 secs

 

[diplomat33]

But it's worth asking whether the Waymo driver can decide for itself where it's safe to pull over and pick up a passenger, or whether the pick up areas need to be pre-labeled by Waymo employees.

I don't know. But if Waymo does label drop off points, I don't think that would be a bad thing. It would make sense to label drop off points to make sure that the car can handle them reliably. You don't want the car to stop in a bad location that would be illegal or unsafe.

 

[diplomat33]

The main issue with Waymo is they don't have Tesla's fleet size and diversity.

Clearly, it is not since Waymo has solved the vast majority of perception cases and has L4 robotaxis in 2 cities, all without Tesla's fleet size or diversity. The lack of a large fleet is not their main issue.

I don't understand anything Waymo is doing nowadays. Just look at what's going on in Chandler (nothing).

As others have pointed out, that's not Chandler. That's the fleet of unused 4th Gen cars in CA because Waymo is now using their 5th Gen cars in CA.

But let me explain to you what Waymo is doing since you don't understand. Waymo is testing FSD on public roads and in simulators, solving edge cases, improving prediction and planning, validating safety, you know, all the stuff you do when you are developing FSD. And right now, they are doing 100,000 autonomous miles per week in SF and doing public testing of a L4 robotaxi service in SF. The next step will be to remove safety drivers in SF and allow full rides for everybody with no NDA. After that, Waymo will undoubtedly expand public testing in a new city. I also expect Waymo to transition all their current robotaxis in Chandler to the 5th Gen I-Pace and retire all the Pacificas.

 

[EVNow]

With HD maps, people underestimate what is being manually labeled and human-verified (by driving and/or double checking vs the images / videos).

I wouldn't be surprised if things like "max speed 3 mph down this driveway" are being manually labeled and/or verified. The car isn't smart enough to figure out practical speeds by itself, and even if it does, it still needs to be human-verified.

There are many many things that need to be manually labeled and/or verified, all traffic control lane semantics as another example.

Not just that - for parking lots - they actually draw the paths that are safe - and if there are any blocks, alternate paths. Imagine scaling that to top 100 US cities.

Unlike what others may say - Waymo has NOT solved general FSD. They have just figured out how to make the car drive around in limited areas by throwing a ton of manual labelers at the problem.

 

[powertoold]

What do Waymo's disengagements in JJRick's videos have anything to do with long range or short range lidars / radars?

It's funny that Waymo is using gen5 in SF when long range lidars don't help at all in hilly SF? Most of your obstacles are within 50 feet.

Waymo's decisions of late don't make much sense, and they don't have much progress to show, especially in Chandler.

 

[Daniel in SD]

What do Waymo's disengagements in JJRick's videos have anything to do with long range or short range lidars / radars?

It's funny that Waymo is using gen5 in SF when long range lidars don't help at all in hilly SF? Most of your obstacles are within 50 feet.

Waymo's decisions of late don't make much sense, and they don't have much progress to show, especially in Chandler.

How much more does a long range LIDAR cost vs. a short range one?
I think you're misunderstanding Waymo's strategy which is to get it working at any cost and then work on bringing the costs down. Sort of like how the original cellphones cost thousands of dollars or the original Roadster cost over $100k.
The obvious issue is that an expensive Robotaxi is worthless since it has to compete with human drivers. So I suspect that Waymo will lose a lot of money for a while. There's also the risk that Tesla will get FSD to work and put them out of business instantly but I doubt they're worried about that.

 

[powertoold]

you're misunderstanding Waymo's strategy

Not sure if anyone understands it at this point.

1) Launch in Chandler ~3 years ago
2) Still hasn't expanded geofence in 2021
3) Failed in dramatic fashion with a single cone (10 feet away) 3 months ago. Roadside assistance has to chase after a rogue Waymo (wut?).
4) Decides to start service in SF??

 

[Daniel in SD]

Not sure if anyone understands it at this point.

1) Launch in Chandler ~3 years ago
2) Still hasn't expanded geofence in 2021
3) Failed in dramatic fashion with a single cone (10 feet away) 3 months ago. Roadside assistance has to chase after a rogue Waymo (wut?).
4) Decides to start service in SF??

I'm not sure what's confusing. They only barely have the system working. Why would they expand to an area that's basically the same as the area they're already in? Just to lose more money? They want to expand to an area that's different so they can improve the system.
Again, it can be the best FSD system in the world and still be crap!

 

[EVNow]

They only barely have the system working. Why would they expand to an area that's basically the same as the area they're already in? Just to lose more money? They want to expand to an area that's different so they can improve the system.

When you fire the top leadership, the new leaders always try to do things differently.

 

[diplomat33]

Not just that - for parking lots - they actually draw the paths that are safe - and if there are any blocks, alternate paths. Imagine scaling that to top 100 US cities.

Unlike what others may say - Waymo has NOT solved general FSD. They have just figured out how to make the car drive around in limited areas by throwing a ton of manual labelers at the problem.

That's a strawman. Nobody is claiming that Waymo has already solved general FSD. Even Waymo is not claiming that they have solved FSD yet. But all their perception, prediction and planning NN are generalized. The only thing that is not generalized is of course is the HD map. The cars use rich semantic information from the HD maps to help guide them but they use generalized NN to perceive the environment and decide what to do.

Not sure if anyone understands it at this point.

1) Launch in Chandler ~3 years ago

Yes, and for 3 years they've run a successful driverless ride-hailing service that nobody has yet to duplicate.

3) Failed in dramatic fashion with a single cone (10 feet away) 3 months ago. Roadside assistance has to chase after a rogue Waymo (wut?).

An embarrassing incident for sure, made worse by mistakes by remote and roadside assistance. But you are exaggerating: it was not 1 cone, it was a closed lane with a series of cones. And let's keep things in perspective: it was one incident in 6M of otherwise safe autonomous driving. We've seen FSD beta fail in much more dramatic and unsafe ways and much more often. But hey, let's keep talking about a cone incident that happened once 3 months ago.

4) Decides to start service in SF??

Yes, because their 5th Gen FSD is good enough and SF is a good market for robotaxis. Waymo's business model is autonomous ride-hailing. So yes, it makes sense for a robotaxi company to start a robotaxi service in a good market like SF.

Funny how you complain that Waymo is not expanding fast enough but when they do expand to a new market, you somehow claim that it does not make sense.

 

[powertoold]

Yes, because their 5th Gen FSD is good enough and SF is a good market for robotaxis. Waymo's business model is autonomous ride-hailing. So yes, it makes sense for a robotaxi company to start a robotaxi service in a good market like SF.

Funny how you complain that Waymo is not expanding fast enough but when they do expand to a new market, you somehow claim that it does not make sense.

My point with gen5 is that it isn't a sensor problem but a software one.

The other point was about Waymo reducing the service area in Chandler, not expanding it. Doesn't show progress at all.

 

[diplomat33]

My point with gen5 is that it isn't a sensor problem but a software one.

The other point was about Waymo reducing the service area in Chandler, not expanding it. Doesn't show progress at all.

I think you are just cherry picking one metric so that you can tell yourself that Waymo is not making progress. You are ignoring other metrics of progress.

For example, Waymo has expanded public testing to SF which is a more difficult environment than Chandler. Being able to do fully autonomous ride-hailing in a more difficult area than Chandler, implies that the software has improved.

Additionally, if you watch the ML presentations, Waymo has shared progress in their perception, prediction and planning NN. So we have concrete examples of how their software has improved.

Lastly, if you look at the safety data, we also progress. From 2019 to 2020, their safety disengagement rate went from 13,000 miles per safety disengagement to almost 30,000 miles per safety disengagement. A jump that big has to be due to software updates.

And Waymo does FSD software updates to their cars on a regular basis.

Frankly, to argue that Waymo has not improved their FSD software much is just silly.

 

[Bladerskb]

At this point powertoold is just knowingly spreading FUD and misinformation.

Here's what he said about a video from Omar in SF "I'm so impressed by V10... There are at least 5 mind blowing decisions and maneuvers in this one..."
Yet that video is inside Waymo's map coverage where they will deploy driverless ride. So either these scenarios are mind blowing therefore they will also be mind blowing when Waymo handles them or they are not. Infact in that very same video a Waymo was seen.

So which is it? Will it be mind blowing when Waymo drives this same route without a driver or is that only reserved for Tesla?

 

[diplomat33]

Carl Quintanilla joins The Exchange from the Code Conference with the co-CEO of Waymo, Tekedra Mawakana. She addresses the trajectory of her company and how autonomous driving can be used to solve today’s labor shortage problems:

A look at the future of autonomous driving

Carl Quintanilla joins The Exchange from the Code Conference with the co-CEO of Waymo, Tekedra Mawakana. She addresses the trajectory of her company and how autonomous driving can be used to solve today's labor shortage problems.

www.cnbc.com

 

[diplomat33]

It's funny that Waymo is using gen5 in SF when long range lidars don't help at all in hilly SF? Most of your obstacles are within 50 feet.

Waymo 5th Gen also has short range lidar. The cars have both short range lidar and long range lidar to best see objects both close and far.

 

[Doggydogworld]

It's funny that Waymo is using gen5 in SF when long range lidars don't help at all in hilly SF? Most of your obstacles are within 50 feet.

Gen4 had long range lidar. The main Gen5 improvement is higher lidar resolution (so high the fanboys were all screaming fake, lol). Resolution helps a lot at long ranges, e.g. you get 20 points for a distant motorcyclist instead of just 2, but it's also great in the city. Especially at night.

Carl Quintanilla joins The Exchange from the Code Conference with the co-CEO of Waymo, Tekedra Mawakana. She addresses the trajectory of her company and how autonomous driving can be used to solve today’s labor shortage problems:

A look at the future of autonomous driving

Carl Quintanilla joins The Exchange from the Code Conference with the co-CEO of Waymo, Tekedra Mawakana. She addresses the trajectory of her company and how autonomous driving can be used to solve today's labor shortage problems.

www.cnbc.com

"What inning are we in?"
"Really great question blah blah.... laser focused blah blah .... and I don't have a date or a timeline...."
Meet the new boss, same as the old boss

 

[diplomat33]

"What inning are we in?"
"Really great question blah blah.... laser focused blah blah .... and I don't have a date or a timeline...."
Meet the new boss, same as the old boss

Not sure I agree with that. The old boss, Krafcik, totally over promised and under delivered. He touted the pre-order of 20,000 Pacificas and apparently told Waymo employees that Waymo would expand to like 9 cities in just 18 months which never happened. I think the new boss is being deliberately evasive on timelines so as not to fall into the same trap of over promising and under delivering (a trap that Elon is very familiar with). Her allusion to Waymo "being a pioneer and being humbled" is basically a polite way of saying "We got really full of ourselves because we thought we were close to solving autonomy and we got bitch slapped by reality". I think she is trying to under promise and over deliver. If she promises nothing then it looks better when Waymo does announce a new expansion or a significant breakthrough.

She also alludes to the difficulty in solving the last 1% of autonomy. That goes to what I was saying before that I think the new co-CEO's have a better grasp of the challenges of solving autonomy than Krafcik did. Waymo now understands that just because you solve a bunch of edge cases and your autonomous driving gets really good, that does not necessarily mean you are close to solving autonomy because there are probably a bunch of new edge cases you did not know about. In fact, the tricky part with solving autonomy is that you don't know where the finish line is because you don't know how many edge cases are left to be solved. You don't know what you don't know. And you will never solve every single edge case everywhere. So, solving autonomous driving is more about reaching a point where it is "good enough" rather than reaching some absolute point where you solved X edge cases.

 

[Doggydogworld]

Not sure I agree with that. The old boss, Krafcik, totally over promised and under delivered. He touted the pre-order of 20,000 Pacificas and apparently told Waymo employees that Waymo would expand to like 9 cities in just 18 months which never happened. I think the new boss is being deliberately evasive on timelines so as not to fall into the same trap of over promising and under delivering (a trap that Elon is very familiar with). Her allusion to Waymo "being a pioneer and being humbled" is basically a polite way of saying "We got really full of ourselves because we thought we were close to solving autonomy and we got bitch slapped by reality". I think she is trying to under promise and over deliver. If she promises nothing then it looks like better when Waymo does announce a new expansion or a significant breakthrough.

She also alludes to the difficulty in solving the last 1% of autonomy. That goes to what I was saying before that I think the new co-CEO's have a better grasp of the challenges of solving autonomy than Krafcik did. Waymo now understands that just because you solve a bunch of edge cases and your autonomous driving gets really good, that does not necessarily mean you are close to solving autonomy because there are probably a bunch of new edge cases you did not know about. In fact, the tricky part with solving autonomy is that you don't know where the finish line is because you don't know how many edge cases are left to be solved. You don't know what you don't know. And you will never solve every single edge case everywhere. So, solving autonomous driving is more about reaching a point where it is "good enough" rather than reaching some absolute point where you solved X edge cases.

You're right that Krafcik initially overpromised. But he just repeated what the engineers told him. I wish I could find the old clip of Dolgov saying self-driving technology was "done". They were outside at some ad-hoc event. It was several years ago, 2018 or maybe early 2019, if anyone has a link.

Anyway, Krafcik did a 180 after his promise of a fully driverless service in 2018 morphed into the pathetic Waymo One "launch". He became very evasive when it came to dates, expansion inside Phoenix, adding cities, etc. Just like Mawakana in this interview almost three years later.

Yes, it's a hard problem. But that's why the get they big bucks. The Chinese are moving faster and will soon pass them.