diplomat33
Average guy who loves autonomous vehicles
Yes, it was. And if there was traffic then traffic would have been blocked until the Waymo moved. We don't know how long the Waymo was "stuck". We also don't know if there was any traffic.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Waymo
- Thread starter Daniel in SD
- Start date
Yes, it was. And if there was traffic then traffic would have been blocked until the Waymo moved. We don't know how long the Waymo was "stuck". We also don't know if there was any traffic.
EVNow
Well-Known Member
And the risk they are willing to take. Google is notoriously risk-averse (just like all rich companies).
If Elon was heading Waymo - he would have pushed harder, and Google would be operating in multiple cities by now.
Seeing the I-Pace Robotaxis all over Palo Alto, you can't help but be struck by how elaborate the LIDAR and camera systems festooned all over the car appear. The I-Paces look really kinda experimentally kluged together and I can't imagine they are inexpensive either, which does appear to be a major advantage of Tesla's non-LIDAR approach.
(Diplomat33, thanks for all the information and your opinions on self-driving. Really, appreciate all your posting !)
diplomat33
Average guy who loves autonomous vehicles
People don't seem understand that the I-Pace is still basically a test car. Yes, Waymo is offering driverless rides in it but it is not the final robotaxi. That is why the sensors looked "tacked on". That's because they are because it is a test vehicle! And yes, the sensors look big and expensive because they are. But they are not the final sensors that will go on the real robotaxi. The Zeekr-Waymo vehicle that was revealed this week will be the real robotaxi. And if you look at it, you can see the sensors are smaller (likely cheaper) and fully integrated into the vehicle. So I am not worried about Waymo's sensors being big or expensive. Again, the I-Pace is a test vehicle, not the "real" robotaxi. Waymo's real robotaxi will be the Zeekr vehicle and the sensors are not big and expensive.
The problem with Tesla's non-lidar approach is that Tesla's FSD is not reliable enough to be driverless yet. It is unproven if Tesla can do driverless robotaxis with their approach. And if you can't do safe and reliable driverless with the non-lidar approach then it does not matter that the sensors are super cheap and small. Now, maybe Tesla is able to achieve safe driverless in a few years from now with their approach (I would be happy if they did) but it would still put them years behind Waymo as Waymo already has safe driverless that they are scaling to new areas. That is why I prefer the Waymo approach because it is proven that it can do safe driverless and the sensors will get smaller and cheaper.
You are welcome.
Last edited:
Do we know if the Zeekr with those sensors will be the last and final Robotaxi version for Waymo? We don't know how well it handles Dense Fog, Freezing Fog, Heavy Rain or even Snow.
Yes, it is a single point of perfection. No further improvement is possible.
Unless Waymo gives up, they will always be a newer robotaxi.
diplomat33
Average guy who loves autonomous vehicles
Will the iPhone 14 be the last and final iPhone? Of course not. Tech is always improving. Waymo is always looking to improve the hardware. I am sure there will be other sensors and other waymo robotaxis with better sensors over time. The Zeekr robotaxi will not be the last robotaxi.
Waymo is working on that. They are doing a lot of testing and validation with the 5th Gen. But even when it does handle dense fog, freezing fog, heavy rain and snow safely, Waymo would still work on even better sensors. You never stop finding ways to make the tech even better. That how the tech industry works.
None of which radar will enable. Will it be Lidar that can see through these inclement weather? Or will it still be a vision call?
I remember diplomat calling it a fusion but in this situation it will be a fusion only if lidar or radar can help see and drive.
diplomat33
Average guy who loves autonomous vehicles
Lidar would probably struggle in those conditions. Cameras can see but with lower visibility. HD radar can see the best through dense fog, freezing fog, heavy rain, and snow. So the Zeekr would probably rely on fusion of radar and cameras to drive in those conditions. To that point, Waymo recently published a paper on their latest radar-camera fusion. They have developed a new neural net dubbed CramNet which handles the fusion of radar and camera data for robust 3D object detection. I believe Waymo is focusing on better radar-camera fusion in order to improve their perception in adverse weather like dense fog, heavy rain and snow.
Here is the paper: Waymo — Research — CramNet: Camera-Radar Fusion with Ray-Constrained Cross-Attention for Robust 3D Object Detection
diplomat33
Average guy who loves autonomous vehicles
I found something interesting. Anguelov showed this diagram of Waymo's autonomous stack. We can see that Prediction and Planning have been merged. They used to be separate in the stack a couple years ago.
Here is the full 30 mn presentation where Anguelov talks about some of their latest ML research:
Here is the full 30 mn presentation where Anguelov talks about some of their latest ML research:
diplomat33
Average guy who loves autonomous vehicles
JJ Ricks interview with Waymo Product Management Director Chris Ludwick!
00:00 Intro and disclaimer
00:19 Start of interview
00:55 Biggest advancements with the 5th gen Waymo Driver?
04:40 What role has rider feedback played in development?
07:15 How do you pick up on indirect feedback from riders?
09:05 How is the 5th gen designed to be more weather resistant?
11:16 What goes into making the riding experience feel so "normal?"
13:37 Favorite edge cases, and how do you deal with them?
17:47 Silly questions begin
18:00 When Waymo merch???
19:03 Dark mode for the Waymo One app
19:27 That weird "our team is working to get you moving" message
00:00 Intro and disclaimer
00:19 Start of interview
00:55 Biggest advancements with the 5th gen Waymo Driver?
04:40 What role has rider feedback played in development?
07:15 How do you pick up on indirect feedback from riders?
09:05 How is the 5th gen designed to be more weather resistant?
11:16 What goes into making the riding experience feel so "normal?"
13:37 Favorite edge cases, and how do you deal with them?
17:47 Silly questions begin
18:00 When Waymo merch???
19:03 Dark mode for the Waymo One app
19:27 That weird "our team is working to get you moving" message
Doggydogworld
Active Member
Hi-def radar will make a huge difference in dense fog and extreme rain/snow. It isn't cheap, though. And the car will still have to slow down (like human drivers) since camera and lidar degrade severely in low-vis conditions.
Because they are.
Cost only matters at scale, and Waymo won't scale the Jaguars.
Waymo said they reduced sensor cost 90% with Gen 4 and cut it in half again with Gen 5. Mass market lidars now cost a few hundred bucks in volume and with the way chip price/performance improves it's almost certain lidars that can match Gen 5 performance will cost that much by the time Waymo scales.
But let's say Waymo-level lidar stays expensive for a few more years. Even $50k of sensor+compute is 10 cents a mile spread over a 500k mile life. Ride hail is $2/mile (or more), so this doesn't really move the needle. It'd be an issue in 2030 when there are enough robotaxis to cause price competition, but I'm extremely confident 100k+ unit prices for sensor+compute won't be anywhere near $50k by then.
diplomat33
Average guy who loves autonomous vehicles
What do you think the Waymo-Zeekr driverless robotaxi will cost? My guess is ~$45k total (vehicle + hardware). My calculation is based on the base vehicle costing ~$20k and the sensors+compute costing ~$25k.
diplomat33
Average guy who loves autonomous vehicles
According to this source, the warranty will be 5 years or 500,000 km.
"The M-Vision will be designed to average more than 16 hours of continuous use per day, according to the developers, and should be able to last at least five years or 500,000 kilometres, according to an announced warranty."
Zeekr launches autonomous SEA-M platform - electrive.com
The premium electric brand Zeekr, part of the Geely Group, has launched its SEA-M platform for autonomous electric vehicles. The SEA-M is an evolution of Geely's Sustainable Experience Architecture (SEA) for autonomous vehicles. The first representative of the new SEA-M vehicle type is the Zeekr M-V
www.electrive.com
Doggydogworld
Active Member
That's in the ballpark. Maybe a bit low on the vehicle -- the smaller BYD Dolphin with 30 kWh LFP pack is $15k in China after subsidy. This will need ruggedized suspension and touch parts for long life in commercial service. And I figure 40 kWh (150-ish range) with 2x per day charging. Instead of a few $K in the various Chinese subsidies you have to add 27.5% US tariff. So probably more like 25k. The smaller Bolt is 26k with 60 kWh NMC pack and without tariff, but that's a blowout price for an end-of-life model.
With enough volume it might make sense for Zeekr and CATL (or whoever) to set up US LFP and vehicle factories to avoid the tariff and grab those sweet Biden subsidies.
Sensor+compute is almost a random number, basically whatever Waymo wants to pay. With solid state or even Ouster-style lidar and Google TPU-based ASICs they could get well below 10k. They aren't the type to skimp, though, so your 25k is probably closer.
diplomat33
Average guy who loves autonomous vehicles
Similar threads
- Article
