Waymo

[EVNow]

Also, remember that we don't have any context. We just have people telling the news media that they are seeing lots of Waymos go down their dead end street. We don't know why. I think we should wait for some official road from Waymo.

May be someone hacked into Waymo

 

[willow_hiller]

Also, remember that we don't have any context.

The video does provide a little context. We can see that for most of the multi-point turns, the car is being operated manually. So we can rule out testing multi-point turns.

And logic would dictate that nothing really changes on a dead-end neighborhood street within a span of 5 minutes. What data could they possibly be collecting at that interval?

 

[diplomat33]

The video does provide a little context. We can see that for most of the multi-point turns, the car is being operated manually. So we can rule out testing multi-point turns.

And logic would dictate that nothing really changes on a dead-end neighborhood street within a span of 5 minutes. What data could they possibly be collecting at that interval?

Who knows? Maybe Waymo is doing imitation learning: collecting steering and accelerator data from manual driving to train the NN on how to handle multipoint turns better?

I just know that it can't be a mislabeled map. Waymo updates the maps automatically when a car sees a problem. So there is no way a mislabeled map would be a problem for that long with cars continuing to get "stuck" week after week.

 

[EVNow]

Maybe Waymo is doing imitation learning: collecting steering and accelerator data from manual driving to train the NN on how to handle multipoint turns better?

Don't they have dedicated testing tracks/area for such things ?

Looks like the drivers needed to improve their safety scores on 3-point turn

 

[diplomat33]

Don't they have dedicated testing tracks/area for such things ?

Yes, Waymo has testing tracks which they use. But that does not mean that they can't also use real world testing as well.

 

[diplomat33]

Here is the issue Tesla is going to have.

From the AI day, my understanding is they use some kind of optimization / heuristics to pick the correct path out of hundreds based on various parameters like comfort, safety, time taken etc. Its difficult to affect particular outcomes compared to if they were using procedural code which used some kind of "if then else" logic to figure out what to do.

@diplomat33 Do you know whether Waymo and others use this kind of optimization techniques as well (my guess would be yes). AFAIK, nobody uses pure imitation learning for path planning.

Yes, I imagine that Waymo probably uses some type of optimization technique. One difference is that Waymo has HD maps that provide additional information to help with the optimization.

 

[powertoold]

The video does provide a little context. We can see that for most of the multi-point turns, the car is being operated manually. So we can rule out testing multi-point turns.

And logic would dictate that nothing really changes on a dead-end neighborhood street within a span of 5 minutes. What data could they possibly be collecting at that interval?

We've seen in Chandler that Waymo likes to use cul de sacs as "pivot" points for their routes. Rather than taking a "riskier" route (usually an unprotected left), Waymo would route cars to do U-turns in cul de sacs. Waymo tries to use suburban / easier routes as much as possible (b/c they need very high reliability for driverless). Here's a recent one in Chandler:

 

[willow_hiller]

We've seen in Chandler that Waymo likes to use cul de sacs as "pivot" points for their routes. Rather than taking a "riskier" route (usually an unprotected left), Waymo would route cars to do U-turns in cul de sacs. Waymo tries to use suburban / easier routes as much as possible (b/c they need very high reliability for driverless). Here's a recent one in Chandler:

That would certainly explain why Waymo seemed unaware of this particular neighborhood; if it's not some planned training exercise, but rather business-as-usual for how they've programmed the routing.

Does make me wonder about the ability for the safety drivers to pass feedback up the chain, though. You think if an issue was getting bad enough for a neighbor to approach a driver and ask what's happening, that driver would have informed their supervisors that something's wrong.

 

[ThomasD]

If it can't turn around can it backup onto the busy street and go the other way? To many cars. Too narrow to turn around. Private drive

 

[Daniel in SD]

I'll post the quick maths on this thread to...

Waymo claims they are test driving 100,000+ miles a week in SF.
There are only 1000 miles of road in SF.
Therefore it is not implausible that they are driving by the same point 50 times in a day.

The cars seemingly not being able to make a u-turn while in self-driving mode does seem like a problem.

 

[sfuape]

I'll post the quick maths on this thread to...

Waymo claims they are test driving 100,000+ miles a week in SF.
There are only 1000 miles of road in SF.
Therefore it is not implausible that they are driving by the same point 50 times in a day.

The cars seemingly not being able to make a u-turn while in self-driving mode does seem like a problem.

I love the mental gymnastics... keep it up!

 

[diplomat33]

Statement from Waymo about the cars going doing the dead end street in SF:

Waymo has sent a statement to Teslarati about the incident. “We continually adjust to dynamic San Francisco road rules. In this case, cars traveling North of California on 15th Ave have to take a U-turn due to the presence of Slow Streets signage on Lake. So, the Waymo Driver was obeying the same road rules that any car is required to follow,” the self-driving company noted.

Waymo’s self-driving cars invade San Francisco street in most bizarre way possible

Something strange is happening in a San Francisco street. As per residents in the area, their Richmond District dead-end street has become the location of some extremely bizarre behavior from Waymo’s self-driving vehicles. Every day, one Waymo after another ends up on the street’s dead-end...

www.teslarati.com

 

[sfuape]

Statement from Waymo about the cars going doing the dead end street in SF:

This is the dumbest explanation.

It's like saying "our av's will never cross the double yellow, NEVER! and neither should you! It's the law!"
LMAO!

Also, seems like their planner has a LONG way to go!...

 

[EVNow]

Statement from Waymo about the cars going doing the dead end street in SF:

Waymo’s self-driving cars invade San Francisco street in most bizarre way possible

Something strange is happening in a San Francisco street. As per residents in the area, their Richmond District dead-end street has become the location of some extremely bizarre behavior from Waymo’s self-driving vehicles. Every day, one Waymo after another ends up on the street’s dead-end...

www.teslarati.com

Where exactly does the law says everyone should go to a dead-end and reverse ?

This is what a "PR" department does. Give inane explanations that hide more than they reveal.

 

[diplomat33]

This is the dumbest explanation.

It's like saying "our av's will never cross the double yellow, NEVER! and neither should you! It's the law!"
LMAO!

It's not like that at all.

This is the road we are talking about. You can see if you drive north on 15th, past Lake St, N 15th St becomes a "slow street" marked with the blue dashed lines on the map.

This is what Google Maps says about "slow streets":

To provide more space between people outdoors, San Francisco’s “Slow Streets” program opens up numerous corridors to pedestrians and cyclists. Some vehicles will also be permitted, though signs will be used to divert traffic and slow overall speeds.

So it makes sense that Waymo would turn around.

 

[diplomat33]

Where exactly does the law says everyone should go to a dead-end and reverse ?

This is what a "PR" department does. Give inane explanations that hide more than they reveal.

It's not a dead end. The street becomes a "slow street":

To provide more space between people outdoors, San Francisco’s “Slow Streets” program opens up numerous corridors to pedestrians and cyclists. Some vehicles will also be permitted, though signs will be used to divert traffic and slow overall speeds.

This just shows how this story is a lot of clickbait and FUD. It's a classic example of the media blowing things our of proportion. Waymo cars are not getting stuck driving down a dead end. They are driving around normally since Waymo is doing a lot of testing in SF. Going North on 15th St eventually become a "slow street" where vehicle traffic is limited. So Waymo decides to turn around. It's a big nothing burger.

 

[EVNow]

So it makes sense that Waymo would turn around.

So, the cars would go to 16th Ave N and then reverse because 15th is a slow street ?

ps :

Going North on 15th St eventually become a "slow street" where vehicle traffic is limited. So Waymo decides to turn around. It's a big nothing burger.

So, they were reversing on 15th itself - if so, that would make sense. But not sure why so many of them in that one road ...

ps : How much area in S.F. are they covering now ... or is this not part of robotaxi service area but just testing ?

 

[Daniel in SD]

Where exactly does the law says everyone should go to a dead-end and reverse ?

The street is a dead end. What I'm confused about is why people think robotaxis shouldn't pick up and drop off rides at dead ends? You can clearly see that there are residences and other cars there!

 

[sfuape]

It's not a dead end. The street becomes a "slow street":



This just shows how this story is a lot of clickbait and FUD. Waymo cars are not getting stuck driving down a dead end. They are driving around normally since Waymo is doing a lot of testing in SF. Going North on 15th St eventually become a "slow street" where vehicle traffic is limited. So Waymo decides to turn around. It's a big nothing burger.

You are wrong, please accept that.
You are so adamant about defending Waymo that you cannot even take a residents word, or use basic tools to review for yourself.

Here is the street view from Google:

 

[diplomat33]

So, the cars would go to 16th Ave N and then reverse because 15th is a slow street ?

No, I don't think so. Based on the Waymo statement, I think they are going up 15th St, cross Lake St, and then when they N 15th St which is a "slow street", they turn around.

Here is the map with an arrow showing the path I think the Waymos are taking.