Waymo

[diplomat33]

Traffic laws are perfect for hard coding, though. ML is imprecise and computationally inefficient. Ideally you've have simple, drop-in traffic law modules for each location. No right on red in Manhattan (and various other places) is literally a couple lines of code, and that code will be 100% reliable. Why maintain a separate neural network for places that don't allow right on red and spend all that time gathering thousands of "don't turn on red" scenarios to train that separate neural net? And still run the risk it might screw up?

The trick is melding the hard-coded traffic law module with generic NNs that handle the bulk of the driving task. No issue for perception and not a big issue for planning, but could be tricky for prediction.

It's stuff like this that make me extremely suspicious of end-to-end claims. Including Musk's.

You raise a good point. It is why I am skeptical of E2E as well. It just seems like a very hard approach. We know road infrastructure is terrible in the US. It is not uniform. And there are big differences in traffic conditions, rules and roads. So training one single NN, with no hard coded rules, to understand the vast differences in rules, traffic and roads in the entire US, seems unlikely. I feel like Elon is attracted to the simplicity and elegance of E2E but it blinds him to the challenges of executing the approach.

 

[hooty]

Probably not since it is outside the scope of the ODD. It requires a lot or prep work in sewing up and assigning a new ODD.

No that is utterly false.
Waymo simply CHOOSES not to take the risks & failures that Tesla does

 

[hooty]

Do they train the entire fleet to drive in all Cities they operate in? Can they take a car that is operating in San Fransisco and drop it in L A and have it navigate around with no major issues? In other words close to human like learning.

It would do it better than Tesla, that’s for certain

 

[Bitdepth]

It would do it better than Tesla, that’s for certain

Can we please not turn this thread into a Vs Tesla thread?

 

[Doggydogworld]

But according to the New York Times article, the fully geared out I-Pace costs as much as $200k.

NYT has no clue and is just guessing, like the rest of us. Krafcik said years ago Gen 5 sensors and compute cost about as much as the i-Pace itself (50-60k) A different time he said the fully outfitted car cost as much as a moderately equipped S Class (bit over 100k at the time). Beyond that we know nothing.

My guess is that the I-Pace is not intended for mass scale, it is intended to be a test and "demo" vehicle.

You're going to make me go dig up your quotes saying Pacifica / Gen 4 was not intended for mass scale, but i-Pace was? Haha.

In 2018 they "ordered" 62,000 Pacificas and 20k i-Paces. They planned to scale both platforms. Sure, Gen 6 would be even better and less expensive, but they weren't going to wait on that. And based on other 2018 statements and actions, they clearly planned to deploy those cars in Phoenix. It's no coincidence that 82k cars is roughly what you need to replace 2nd / 3rd cars there.

Their plan failed. Utterly and completely. They built it, but no one came.

Waymo has great technology and a vague sense of market needs. But they have no entrepreneur / marketing genius types to drive them forward. So they just drift along. End game? Who knows. Their tech lead will diminish over time. My bet is they'll eventually get passed, just like Xerox PARC.

 

[hooty]

Can we please not turn this thread into a Vs Tesla thread?

Fair enough. Its better to compare companies who actually have or will have in the next few years robotaxis

 

[diplomat33]

You're going to make me go dig up your quotes saying Pacifica / Gen 4 was not intended for mass scale, but i-Pace was? Haha.

In 2018 they "ordered" 62,000 Pacificas and 20k i-Paces. They planned to scale both platforms. Sure, Gen 6 would be even better and less expensive, but they weren't going to wait on that. And based on other 2018 statements and actions, they clearly planned to deploy those cars in Phoenix. It's no coincidence that 82k cars is roughly what you need to replace 2nd / 3rd cars there.

I acknowledge that Waymo had planned to scale big in 2018 and that did not work out. At the time, it looked like they would scale but then things changed. Obviously, current plans to scale the Geely vehicle could also fail. We won't know until it happens or does not happen.

And I could be wrong when I say that the I-Pace won't scale. Waymo had said even recently that they would scale with the I-Pace. So maybe Waymo will try to scale big with the I-Pace. Of course that was before the Jaguar announcement about discontinuing production of the I-Pace. I just think it makes more sense for Waymo to scale big with the cheaper Geely vehicle.

Waymo has great technology and a vague sense of market needs. But they have no entrepreneur / marketing genius types to drive them forward. So they just drift along. End game? Who knows. Their tech lead will diminish over time. My bet is they'll eventually get passed, just like Xerox PARC.

Waymo is not stagnant, they are making tech progress all the time. So I think they will keep their tech lead for years to come. But their tech lead may narrow. I think it would be a real shame if Waymo squandered their tech lead. My concern is not that Waymo will lose their tech lead but more than they won't scale fast enough. I still think that even if Waymo fails to expand robotaxis at scale that they can still leverage their tech somehow. They are not going to just throw away decades of research. So worst case scenario, if robotaxis don't pan out, I imagine that they could adapt the Waymo Driver to do some form of autonomy on consumer cars or something. All the software would still be good after all.

 

[diplomat33]

00:00 Car arriving
00:16 Boarding
00:27 Ride start and parking lot navigation
01:29 VERY impressive unprotected left full road crossing
02:05 Unexpected braking
03:14 Construction and cones on right side
04:45 End of construction
05:09 Right lane change
06:30 Zoom sucks
06:40 Rider screen fog of war demo
08:00 Protected right
08:10 Left lane changes
11:33 Unprotected left
12:10 15mph school zone
13:10 Four way stop right turn
13:40 Turn into parking lot

 

[enemji]

Hey @diplomat33 - How good is the Waymo at handling potholes?

 

[diplomat33]

Hey @diplomat33 - How good is the Waymo at handling potholes?

I don't think we've seen any videos of Waymo encountering potholes. So I can't really say one way or the other. I will ask Maya who is a well known Waymo rider who has done hundreds of rides in SF and see if she can provide an answer.

 

[diplomat33]

Hey @diplomat33 - How good is the Waymo at handling potholes?

Does not directly answer about potholes but here is Maya's answer:

https://twitter.com/x/status/1695193106973372471

 

[enemji]

Does not directly answer about potholes but here is Maya's answer:

https://twitter.com/x/status/1695193106973372471

Thsnks! So it seems that is still something to work upon. My FSDb 11.3.6 seems to handle itself very well with cars parked on both sides of the street side. It takes a path over the mid line marker and even dunks behind one when a vehicle is approaching from the other side. But potholes is still something it goes right into/over.

 

[diplomat33]

By the way, Waymo confirms everything runs in the car. The cars are equipped with a cellular connection to get real-time road conditions but it is not needed to drive:

"We believe the most optimal way to operate autonomous driving technology is for the compute to be on board and for it to make decisions, without needing to rely on cell signals and remote operators," the spokesperson wrote. "We don't want to have a situation where, say, if the car lost cell connection, it couldn't make a left turn. Therefore, everything from custom made maps to robust neural nets that inform our perception and routing runs on board."

"Waymo's vehicles are equipped with a cellular connection, which is used to supplement the autonomous driving system with real-time road conditions, said the spokesperson"

https://www.lightreading.com/iot/waymo-keeps-5g-way-in-background/d/d-id/786283?_mc=RSS_LR_EDT

 

[diplomat33]

Nice driverless from Golden Gate bridge to Fisherman's Wharf.

 

[Bitdepth]

Nice driverless from Golden Gate bridge to Fisherman's Wharf.

I remember when the argument was waymo was scared to drive in a dense urban populated area, now they're just handling it like nothing.

 

[diplomat33]

I remember when the argument was waymo was scared to drive in a dense urban populated area, now they're just handling it like nothing.

Yeah, what I like seeing is that Waymo is safely handling a more diverse ODD now than when they started in Chandler. Waymo drives safely in dense urban driving but also in more arterial roads like we see in this video. And apparently, it handles highways very well too (although we have not seen driverless highway yet). We've also seen videos of Waymo driving safely in some moderate rain. So we are seeing Waymo safely drive in a lot of different environments which bodes well for a generalized driver.

 

[diplomat33]

New JJ Ricks video!

00:00 Begin
00:35 Unprotected right
02:30 Little bit of unexpected braking
02:50 Unprotected right, busy road
03:05 Freaky butt-clenching forced merge (but still very good. title of the video here)
04:20 Left lane change
05:30 Little bit of a rocky start
06:40 Left lane change
08:01 Protected left
08:11 Windshield wipers go off
08:48 Pedestrian freaks the car out a bit
09:12 Detour right turn into neighborhood to avoid right turn at stoplight
09:48 Unprotected right
11:25 Protected left, welcome to ASU
12:20 Long pointless detour
12:40 Tight squeeze
14:19 Weird path
15:13 Stop sign left
15:40 Protected right, end of detour
16:06 Interesting cul-de-sac

 

[Doggydogworld]

JJ said he almost never sees Waymo testing on highways any more. Doesn't bode well for "highways are imminent" theory.

For a while Cruise was forcing Waymo to pick up their pace. But DMV dialed Cruise back 50% so Waymo can now relax. Welcome back to tortoise mode.

 

[diplomat33]

JJ said he almost never sees Waymo testing on highways any more. Doesn't bode well for "highways are imminent" theory.

Or they don't need to test anymore because they finished highway validation. Or they are testing on highways in other cities now and not around Phoenix anymore because they finished their testing on Phoenix highways.

 

[diplomat33]

For a while Cruise was forcing Waymo to pick up their pace. But DMV dialed Cruise back 50% so Waymo can now relax. Welcome back to tortoise mode.

With the new president of Alphabet coming in and cracking the whip, I doubt Waymo will enter tortoise mode.