....but watching someone march one step ever 12 hours is kinda disheartening and hard to actually tell if they ARE marching or standing still. After all these years they need to pick up the pace a little and hit 3 to 5 more cities this year to start testing.
-
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
I'd say they are done "testing" in southern climates. Now it's just about scaling out carefully not to "Cruise" the brand. The LA rollout was what? Free rides for one month, then paid.
This is a weird failure though, from SF:
You don't see too many of these anymore though.
diplomat33
Average guy who loves autonomous vehicles
It is a weird failure for sure. My guess is the Waymo thought the adjacent red light for the other lane was for its lane so it assumed it was doing everything right to stop for the red light. Obviously, Waymo has multiple redundancies for checking what traffic light is for which lane. It is a reminder that ML is never 100% reliable. Even well trained ML and redundancies like HD maps, will still produce failures. It is just a matter of how rare the failures are and how serious the failures are. The goal of adding redundancies is not to eliminate failures but to make the errors super rare. This error seems pretty rare for Waymo considering it only seems to happen every couple million miles.
diplomat33
Average guy who loves autonomous vehicles
Waymo is scaling up number of rides and passengers quite a bit in SF, according to CPUC data:
"Waymo robotaxis logged 1.4 million driverless ride-hailing miles between December and February, roughly a 42% increase from the prior quarter, according to data the company reported to the California Public Utilities Commission.
December was the first month Waymo reported carrying more than 100,000 passengers on driverless trips — the vast majority of which were on paid rides — and the Alphabet-owned company reported similar figures for January and February.
That quarterly haul of about 316,000 passengers was up 45% from the quarter covering September through November, when Waymo carried about 218,000 riders. The latest figures represent a huge jump from this time last year, when the company’s robotaxis shuttled about 13,000 people on unpaid rides."
So they are not exactly standing still IMO. But right now, they are focusing on building up number of rides in existing cities rather than expanding to lots of new cities. It is a deliberate choice to go "slow" in terms of number of new cities. Waymo seems to want to focus on quality over quantity. They prefer to build up the service in a city first before moving on to the next city.
Here is what Tekedra said recently at SXSW:
"We could go to every major city in the US, plant a flag, and stand up a service and compete with the local options that are available. The reason that we take the approach that we're taking, is we want to build a full service. We want to see that it's actually integrated into people's lives. And then we'll move to the next city."
diplomat33
Average guy who loves autonomous vehicles
Waymo starting paid rides in LA today:
50k people on waitlist:
www.nbcnews.com
50k people on waitlist:
Waymo will launch paid robotaxi service in Los Angeles on Wednesday
Autonomous taxis are getting more buzz, as GM subsidiary Cruise said it was resuming some operations and Tesla CEO Elon Musk said he was planning a product launch in August.
www.nbcnews.com
Doggydogworld
Active Member
Yes, very weird. The visualization shows a red traffic light at 3:40. It switches to green at 3:47 then disappears after a few seconds. Then a red stop sign symbol appears out of nowhere! There is no actual stop sign I can see, not even a hand-held one or a T-shirt or store ad or something.
The stop sign disappears when Waymo reaches in the crosswalk, but that's just how they do visualizations now. You see this at 1:06, they show a stop sign to explain why they're stopping. It disappears when they come to a stop at 1:09 even though the physical stop sign is obviously still there. It's the same with red/green lights, they only show them temporarily.
I don't think it was confused by the traffic lights for the lanes to the right. The visualization would have shown a red traffic light instead of a stop sign. And Waymo would have proceeded when those lights turned green. Finally, their pre-mapping and pre-driving has to be better than that.
My wild guess is the moped rider ahead moved his white helmet just enough at 3:52 to make part of the red truck in the distant foreground look like a stop sign. The effect lasted a while even after the hallucination vanished due to object persistence logic. Eventually the persistence timed out or remote ops overrode it.
diplomat33
Average guy who loves autonomous vehicles
The top of the red truck does not look like a stop sign. But even if the car did somehow think the red truck was a stop sign, it would be odd for the effect to last 3 minutes. Plus, we see the Waymo stop for the light train that passed and then still wait a bit before going. I don't think the hallucination would last that long. And if we look at some JJ Ricks videos, visualization hallucinations usually only last a few seconds, not several minutes. So again, I think it would be odd for a perception error to last that long.
But the message on the screen said "waiting for intersection to clear" which would imply it thought there was something in the intersection that would be in the way. (But it didn't visualize anything.)
Doggydogworld
Active Member
Your chart proves my point. Waymo grew very fast from the 10x/year statement until October. Then they slowed drastically to a 2.5x/year rate. I wonder what could have happened in October? Oh yeah, their only competitor self-vaporized. The pressure to grow is gone.
AlanSubie4Life
Efficiency Obsessed Member
This is why you should never rely on lidar, which is a crutch.
diplomat33
Average guy who loves autonomous vehicles
You know Waymo does not rely on lidar only. Waymo also uses camera vision, radar and HD maps. Come on now! I expect better from you!
Last edited:
AlanSubie4Life
Efficiency Obsessed Member
I firmly believe I should not need to put
on every single post.
diplomat33
Average guy who loves autonomous vehicles
According to teslafsdtracker, the intervention rate is 146 miles per safety critical intervention. Tesla needs to get to at least 10k miles per safety intervention to even start a geofenced robotaxi. Put differently, 96% of drives have no safety critical intervention. Tesla would need to get that number to like 99.999% to do robotaxis. V12.3.3 is a very long way from eyes-off or driverless. The notion that vision-only is almost solved is silly.
IMO, there is no need to remove radar/lidar yet. Cost of radar and lidar are coming down and radar and lidar add huge safety benefits. Comparing stats, Waymo is at 99.999959% in terms of reliability while Tesla FSD V12 is at 96%. So using radar/lidar, Waymo's autonomous driving is way more reliable than Tesla's vision-only. Why would Waymo remove something that is getting cheaper and makes their autonomous driving orders of magnitude more reliable?
Last edited:
It's nice he takes the time to put it together but I wouldn't reference anything cybrlft generates for FSD stats. Much is the same for TSLA provided stats. And until there's more visibility there probably won't be a reliable source.
diplomat33
Average guy who loves autonomous vehicles
It's the best we got since Tesla won't release official, verified, intervention data.
Similar threads
- Article
