Waymo

[Bladerskb]

Maya on Reddit shared some examples of Waymo handling temp stop signs and construction:

temp stop sign, downed tree:

temp stop sign, lots of cones:

construction/struck cruise:

There are now hundreds of clips of driverless Waymo handling construction, road closure and thousands of clips of it handling double parked cars. Yet many people who will ignore it and say:

"I don't think their approach is going to work well, as it partially relies upon HD maps. The Achilles heel of HD maps is that they go out of date almost instantly (what happens when a car double-parks? That wasn't on the map). What happens if there is construction ongoing with lane shifts since the mapping?"

@bkp_duke

 

[EVNow]

There are now hundreds of clips of driverless Waymo handling construction, road closure and thousands of clips of it handling double parked cars. Yet many people who will ignore it and say:

"I don't think their approach is going to work well, as it partially relies upon HD maps. The Achilles heel of HD maps is that they go out of date almost instantly (what happens when a car double-parks? That wasn't on the map). What happens if there is construction ongoing with lane shifts since the mapping?"

@bkp_duke

Why do they need HD Map if they can handle areas that look nothing like their HD map ?

Why are they not operating in my area ?

 

[bkp_duke]

Why do they need HD Map if they can handle areas that look nothing like their HD map ?

Why are they not operating in my area ?

BINGO. Winner winner chicken dinner.

Even if they can handle double-parked cars (the one I was in SF was not able to do this back in April), the "value" of this system is still incredibly low due to the teeny tiny geo-fenced area it is confined to.


EDIT - I got FSDb 11.4.2 last night, had 3 interventionless drives, one that I intervened (4th drive), but am not sure it would have had a safety problem in the one instance I took over or not, I was being cautious. Drove on the highway, drove in the suburbs, and went to go get tacos and donuts (not back to back, ouch on my GI tract). The driving was very competent and very smooth overall. Not perfect, but a marked step forward from 11.3.6.

 

[ThomasD]

This is just a picture. But I got stuck behind one of these today and wondered if Waymo encountered a vehicle like this would it try to drive under it?

 

[Bladerskb]

BINGO. Winner winner chicken dinner.

Even if they can handle double-parked cars (the one I was in SF was not able to do this back in April), the "value" of this system is still incredibly low due to the teeny tiny geo-fenced area it is confined to.

(moderator edits) Easily debunkable … by thousands of videos at ... Precisely as I predicted. Its gonna be a rough next couple years for the Tesla faithful. This is definitely the second best.
I think the best was one by someone saying "Waymo is just a glorified AEB, nothing more." Some others are "Tesla uses NN to predict what other pedestrians and cars will do, others don't even know what a NN is".

Gotta love Tesla fans.

 

[Bladerskb]

Why do they need HD Map if they can handle areas that look nothing like their HD map ?

Why are they not operating in my area ?

Why drive with two eyes when you can drive with one?
Why two ears when you only need one?
Why two kidneys?
Why multiple load balancing server?
Why backup compute?
Why do websites use backup servers?
Why do airplane have backup of *every component?

This is like grade level questioning, which I know you are beyond.
This is pure example of bad faith.

Because the same type of person who says "Tesla FSD will be way better if it had HD map like Waymo" will then turn around and say "Why do Waymo need HD Map if they can drive on the fly".

Its a complete waste of time.

 

[HVM]

There are no fields or need of spraying in any of the Waymo areas. (But I bet that Beta would folds mirrors and try to drive under it, if the sprayer was stationary...)

 

[bkp_duke]

It wonderful that Tesla fans have now resulted to actual lies as they have nothing else to cling to. An easily debunkable lie by thousands of videos at that. Precisely as I predicted. Its gonna be a rough next couple years for the Tesla faithful. This is definitely the second best.
I think the best was one by someone saying "Waymo is just a glorified AEB, nothing more." Some others are "Tesla uses NN to predict what other pedestrians and cars will do, others don't even know what a NN is".

Gotta love Tesla fans.

Ah yes, the last resort - the personal attack.

Why do I need to lie? Waymo is geofenced to 0.0001% of the streets that FSDb can cover, it's slower in going from point to point (by more than 2X), and is accepted as a "solution" that simply cannot scale. The DATA STORAGE requirements alone for HD mapping to cover all of the USA is Petabytes of data. Where are you going to store that in the car? They can't be transmitted on the fly, the network latency for LTE and 5G is too slow to pull that kind of data in realtime.

But hey, you do you. Clutch to your emotional arguments over facts.

 

[Bladerskb]

Ah yes, the last resort - the personal attack.

Why do I need to lie? Waymo is geofenced to 0.0001% of the streets that FSDb can cover, it's slower in going from point to point (by more than 2X), and is accepted as a "solution" that simply cannot scale. The DATA STORAGE requirements alone for HD mapping to cover all of the USA is Petabytes of data. Where are you going to store that in the car?

But hey, you do you. Clutch to your emotional arguments over facts.

Zero Personal Attack.

There are hundreds/thousands of video of Waymo doing something.
There are hundreds/thousands of video of Tesla doing something.

Person A who is a Tesla Fan: Tesla can do it, look at all the videos. But Waymo clearly can't.
Person B who is not a Tesla fan: Tesla and Waymo can both do it based on the clear evidence.

Person B says to Person A: Waymo can clearly do it. look at all the videos just like you said of Tesla, to say otherwise despite the clear evidence is a balant lie!
Person A: That's an emotional argument!

This is the problem with Tesla fans, you can't see past your bias.

 

[EVNow]

This is pure example of bad faith.

Oh shut up.

You know it is a very difficult question to answer - so you are absolutely BSing.

Again - if you can drive without HD Map - why use HD maps ? Afterall if you don't need HD Map - you can use the "very great" software that has "already cracked FSD" to drive autonomously anywhere in the world ! Isn't that great ....

BTW, you have conflict of interest that you have never explained. You are here as a "guest" - so don't abuse the hosts. Be nice. Don't indulge in personal attacks.

 

[bkp_duke]

Zero Personal Attack.

There are hundreds/thousands of video of Waymo doing something.
There are hundreds/thousands of video of Tesla doing something.

Person A who is a Tesla Fan: Tesla can do it, look at all the videos. But Waymo clearly can't.
Person B who is not a Tesla fan: Tesla and Waymo can both do it based on the clear evidence.

This is the problem with Tesla fans, you can't see past your bias.

No bias, I've used all these platforms (FSDb, Waymo, Cruise). Waymo and Cruise are slow in point to point drives, and have in general a solution that simply doesn't scale.

GREAT if you have experienced them going around double-parked cars. Perhaps the road I was on in SF was too narrow for it to attempt that. But the FACT remains I was sitting in the damned car, not you, and it would not attempt to go around a double-parked UPS truck making a delivery.


Food for thought - how many of those instances of going around a double-parked car you report are actually the software doing it, vs. someone remotely? We don't know the answer to that, do we?

 

[EVNow]

GREAT if you have experienced them going around double-parked cars. Perhaps the road I was on in SF was too narrow for it to attempt that. But the FACT remains I was sitting in the damned car, not you, and it would not attempt to go around a double-parked UPS truck making a delivery.

It’s nonsensical to say just because in some video Waymo or FSD did something- it will always do it correctly and if someone says it didn’t, they are lying.

If we use that “logic”, every time someone writes FSD did something wrong, we can say they are lying because there is a video of FSD doing it correctly.

 

[bkp_duke]

It’s nonsensical to say just because in some video Waymo or FSD did something- it will always do it correctly and if someone says it didn’t, they are lying.

If we use that “logic”, every time someone writes FSD did something wrong, we can say they are lying because there is a video of FSD doing it correctly.

I'm assuming you are referring to the other poster, because I was reporting MY OWN personal experience with double-parked cars on my Waymo ride in SF back in April.

 

[Doggydogworld]

Why do they need HD Map if they can handle areas that look nothing like their HD map ?

It adds another couple 9s of safety. Not needed for a consumer toy, but important for driverless operation.

Why are they not operating in my area ?

They don't like you.

Seriously. Will you spend 5k a year with them? And your neighbors, too? No? Then you don't matter. They'll start with low hanging fruit.

Uber does 20m+ rides per day. And some crazy high fraction are in just 5 global cities. Waymo is on track for 20k/day around September 2024. Even at their 10x/year growth rate they'll still only be 1/10th Uber's size in Septeimber 2026. You won't matter until then, at least. And 10x/year is insanely fast, it's not at all clear they can sustain that.

 

[EVNow]

It adds another couple 9s of safety. Not needed for a consumer toy, but important for driverless operation.

So, when HD Map is not available they are not as safe ? Do Waymophiles agree?

 

[rbt123]

So, when HD Map is not available they are not as safe ? Do Waymophiles agree?

Not sure what a Waymophile is, but yes, that's true of any automated vehicle or human driver.

Humans also have the most accidents in areas they're least familiar with too (on a per mile driven basis). Having a high certainty of what is behind visual obstructions (like transport trucks, garbage trucks, school buses, etc.) is highly beneficial. Knowing what is around a blind corner (bike lane ends, so they merge into general traffic) helps improve predictions of what other road users will do.

A decent HD map will include not just physical items, but behavioural too: I.e. at 5pm most eastbound vehicles turn left to northbound at intersection X/Y. That information helps improve accuracy of behaviour predictions, or drivers often abruptly swerve around a pothole 50m down the street so give them some space. Tesla should have a massive amount of behavioural information about other drivers at various times of day and various seasons which, if everything else was equal, would give them a safer ride than Waymo.

To ignore this hard gained knowledge to improve predictions and fill out (with high probability of correctness) visually obstructed areas isn't wise. Accurate data nearly always leads to better machine learning decisions. Even marginal safety gains in driving can be worth tens of millions in reduced insurance claims over a very large fleet.

 

[EVNow]

Humans also have the most accidents in areas they're least familiar with too (on a per mile driven basis).

Ironically people have most of the accidents within a mile, IIRC, of their homes.

And people don't have cm level memory of all the areas they drive in.

 

[EVNow]

Seriously. Will you spend 5k a year with them? And your neighbors, too? No? Then you don't matter. They'll start with low hanging fruit.

Uber does 20m+ rides per day. And some crazy high fraction are in just 5 global cities. Waymo is on track for 20k/day around September 2024. Even at their 10x/year growth rate they'll still only be 1/10th Uber's size in Septeimber 2026. You won't matter until then, at least. And 10x/year is insanely fast, it's not at all clear they can sustain that.

Seriously, its not here because they can't handle rain (and occasional snow - unlike occasional Sun).

Thats why they are not operating here - even though they do run tests here. Uber & Lyft had 3.3 million rides in Q1 of '22. I'm sure its higher than in Chandler, AZ

 

[powertoold]

When the discussion gets this heated, I just say let's revisit this in 6 months If Waymo and Cruise are dead ends, we should know by the end of this year for sure.

 

[EVNow]

When the discussion gets this heated, I just say let's revisit this in 6 months If Waymo and Cruise are dead ends, we should know by the end of this year for sure.

I doubt it. Unless some external force compels them - all AV projects will continue. No reason to stop unless finance dries up.