Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Waymo

This site may earn commission on affiliate links.
You guys are comparing a mass production vehicle to a specialized vehicle built for one company? If Tesla only built 10,000 Model 3s, each of them would probably cost $150,000.
I have a Lotus Elise. They built ~5000 per year of those, in England no less, and retail price was $45k in the US.

GM builds ~25k Corvettes a year on average. The base Origin vehicle has more in common with a golf cart than a Corvette. I don't know if it will even be highway capable. So far Cruise doesn't exceed 25 mph.
 
The base Origin vehicle has more in common with a golf cart than a Corvette. I don't know if it will even be highway capable. So far Cruise doesn't exceed 25 mph.

I believe there is a federal regulation that limits vehicles with no steering wheels to 25 mph, unless they get a special exemption. So the Origin will probably be restricted to 25 mph. Plus, the Origin seems designed to basically serve as a city shuttle. So, it will probably just do ride sharing around the city at 25 mph. I don't think it is ever intended to drive on the highway.
 
I believe there is a federal regulation that limits vehicles with no steering wheels to 25 mph, unless they get a special exemption. So the Origin will probably be restricted to 25 mph. Plus, the Origin seems designed to basically serve as a city shuttle. So, it will probably just do ride sharing around the city at 25 mph. I don't think it is ever intended to drive on the highway.
Looks like Cruise is not planning to limit it to 25mph. Here is the NHTSA "Notice of receipt of petition for temporary exemption; request for public comment."

"NHTSA seeks comment on whether the agency should apply the same types of conditions that it applied to Nuro's exemption for ADS-equipped low-speed occupantless vehicles. NHTSA seeks comment not only on whether these conditions are appropriate to apply to GM's exemption, if granted, but also whether there are additional terms that NHTSA should apply. GM's exemption request differs significantly from Nuro's in that the request is for a passenger vehicle, and it is not limited to 25 mph, as was the case of the Nuro vehicle. As such, there are likely to be additional terms that would be appropriate to apply to GM's exemption, if granted."

Cruise Origin weighs 3,084kg! This is probably a good idea for passenger safety given how often people run into them.
 
Looks like Cruise is not planning to limit it to 25mph. Here is the NHTSA "Notice of receipt of petition for temporary exemption; request for public comment."

"NHTSA seeks comment on whether the agency should apply the same types of conditions that it applied to Nuro's exemption for ADS-equipped low-speed occupantless vehicles. NHTSA seeks comment not only on whether these conditions are appropriate to apply to GM's exemption, if granted, but also whether there are additional terms that NHTSA should apply. GM's exemption request differs significantly from Nuro's in that the request is for a passenger vehicle, and it is not limited to 25 mph, as was the case of the Nuro vehicle. As such, there are likely to be additional terms that would be appropriate to apply to GM's exemption, if granted."

Cruise Origin weighs 3,084kg! This is probably a good idea for passenger safety given how often people run into them.

Thanks. So Cruise is seeking an exemption. Good to know.
 
Thanks. So Cruise is seeking an exemption. Good to know.
They applied for an exemption for the Bolt back in 2018. They pulled that application in 2020 when they revealed the Origin and said they'd seek an exemption for it, instead.

Cruise is permitted for 30 mph in San Fran but I never see them go above 25. I think that's intentional, to figure out the market for a 25 mph service in case NHTSA doesn't approve their Origin exemption in time.
 
  • Like
Reactions: diplomat33
Wow. Looked like their were a few opportunities before it finally went. FSDb and Waymo UPLs might be a tie.

Yeah there were a couple opportunities where the Waymo could have probably tried it, especially if was assertive enough. There was even one moment where it looked like a car was stopping to let the Waymo go. But ultimately, the goal is to do the UPLs consistently safe. So if it means waiting a bit longer and ensuring a safe turn, that's probably better.

One thing that made this UPL tricky IMO is that it was not always easy to see the cross traffic because there were some objects and cars blocking the view. Making the turn at the wrong time could mean pulling out in front of traffic you did not see. Also, it was at night where you could see headlights but not always see the actual vehicle. I don't think we've seen Chuck attempt UPLs at night like this, it is always during the day with good visibility. Overall, I think the Waymo handled a difficult UPL (fast moving traffic, blind spots, poor visibility) like a champ. It creeped forward to aid in visibility and made the turn when it was safe.

I would add. I know there has been a lot of talk that Waymo reroutes to avoid all UPLs because they are too afraid of even the slightest difficult turn. So the fact that Waymo let's the car attempt this UPL with no safety driver, says a lot about their confidence IMO. It also shows that Waymo is definitely letting the 5th Gen do UPLs and the 5th Gen can handle them.
 
Last edited:
FSDb and Waymo UPLs might be a tie.
I’d say FSDb has the edge here, for sure. But of course you have a human at the wheel to squash any of those fatal attempts. So hard to compare.

The occupants were amused, apparently at how bad it was. Not surprisingly.

Pretty sure all robotaxis are kind of garbage, though one of these days I’d like to ride in one so I can stop speculating.
 
  • Funny
Reactions: diplomat33
I’d say FSDb has the edge here, for sure. But of course you have a human at the wheel to squash any of those fatal attempts. So hard to compare.

The occupants were amused, apparently at how bad it was. Not surprisingly.

Pretty sure all robotaxis are kind of garbage, though one of these days I’d like to ride in one so I can stop speculating.

I would give you a disagree but I know people hate that. But I disagree with everything.

FSDb does not have the edge. Waymo waited awhile to complete the turn but there was good cause to wait. There was a lot of traffic and it was dark, with obstacles blocking the view. In contrast, I've seen FSDb in my own car, just sit there a long time and slowly creep forward to do a simple turn during the day, with no traffic at all. And if you need a driver to squash the fatal attempts, you automatically fail. Waymo handled this difficult UPL in the dark with a lot of traffic without a human driver. I would say Waymo handled a difficult scenario better than FSDb handles a simple scenario. So Waymo has the edge.

They were not amused at how bad it was. They joke at first that they will be there awhile but they were both super impressed at the end when it completes the turn. JJ even tweets that he probably would have avoided the turn altogether. So Waymo handled driverless a turn that a human would not have attempted!

Calling all robotaxis "kind of garbage" is unfair. AVs already handle a lot of driving really well. In fact some AVs can handle some situations better than human drivers. But yes, there are some situations where they are more cautious than a human. That does not make them "garbage".
 
Last edited:
I would add that AVs are potentially much better than humans at deciding when to complete a maneuver like a UPL. That's because AVs have better data than humans. AVs like Waymo, calculate the precise position and speed of every object with cm level precision. And the computer is doing trillions of calculations per second, calculating the most probable path of hundreds of objects to 8 seconds into the future. Humans don't actually know the precise speeds or distances of objects on the road. Instead, humans measure relative speed. We know if an object is getting closer or further way. We know if an object is accelerating or slowing down. We can guesstimate the time it will take to complete a maneuver like a UPL in time. But humans can get it wrong and when that happens, we can cause accidents. There are accidents all the time where a human driver tries to do a UPL, thinking they have enough time, but they miscalculated or another car changes speed and they can't make it and they hit the other car or another car hits them. AVs are much less likely to make those errors of misjudging distance or speed of other vehicles. Of course, having better data does not necessarily mean the AV will always make better decisions, but it helps a lot.
 
FSDb does not have the edge. Waymo waited awhile to complete the turn but there was good cause to wait. There was a lot of traffic and it was dark, with obstacles blocking the view


I mean with awesome LIDAR it should have been no problem but missed huge gaps you could sail an aircraft carrier through at around 1:11 remaining and :26 remaining. These were more than 5 second gaps. Just go! The car knows nothing is coming - as you say it has super human perception - so what is the problem? I could see these gaps with my eyes - you can tell from reflections and stuff that there is no traffic coming. Trivial perception job.

And then in the end it sits in the middle of the street.

Just not impressive.

And yes FSD would have been awful most likely. Hard to know on any given day; sometimes it just YOLOs it. I said it was hard to compare since one has a safety driver.

And if you need a driver to squash the fatal attempts, you automatically fail.

Yeah, not comparable, at all.
 
I mean with awesome LIDAR it should have been no problem but missed huge gaps you could sail an aircraft carrier through at around 1:11 remaining and :26 remaining. These were more than 5 second gaps. Just go! The car knows nothing is coming - as you say it has super human perception - so what is the problem? I could see these gaps with my eyes - you can tell from reflections and stuff that there is no traffic coming. Trivial perception job.

The problem is not perception. The Waymo has superhuman perception. It sees the gaps perfectly with no problem. But Perception just provides data to the Planner, it does not decide when to go. The Planner makes a decision about when to go. Perhaps the Planner thought the gaps were not big enough? Or maybe the Planner was not confident enough in the prediction about what the other vehicles would do? In any case, the Planner made the decision when it was safe enough to go. At about the 2:30 mark when there is a big gap, the car does go. It yields for that one white truck passing and then does make the turn confidently. I guess you think it should have gunned it in the turn like FSDb does. I think Waymo does not like to do that because sudden accelerations can be unpleasant for passengers. But maybe the Planner needs some more training to handle this type of scenario more assertively when there is a big gap. But that is a planner issue, not a perception issue.
 
Last edited:
  • Like
Reactions: AlanSubie4Life
. I guess you think it should have gunned it in the turn like FSDb does. I think Waymo does not like to do that because sudden accelerations can be unpleasant for passengers.
No, just want it to drive like an excellent human driver. Confident, assured, comfortable ride, with minimized jerk.

FSDb is not a good model for this.
 
  • Like
Reactions: diplomat33
Waymo ride (with safety driver) in the rain with lots of situations, including several construction zones, rerouting through parking lot and one disengagement when the car was slow to respond to a cross guard waving the car forward, and a "computer issue" that prevented re-engagement of the autonomous driving.


0:00 Introduction
0:17 Start of Ride
0:51 Passing a road closure sign
1:11 Right turn with pedestrian in crosswalk
3:12 Driving through intersection with cones
3:46 Lane closed by cones
4:07 Taking right of way at a stop sign
4:31 Pullover for waypoint
5:28 Creeping for unprotected intersection
6:12 No reaction to to flying birds
6:18 Slowing for pedestrian in street
6:29 Slowing for right of way violator
7:05 No reaction to flashing work truck
7:28 Construction mayhem begins: lane shift
8:14 Snaking through the parking lot
9:43 Remote assistance is watching
10:52 Seeing Machines driver monitoring
12:20 Passing a vehicle waiting to make a turn
13:10 Inactive emergency vehicle
13:52 Slow to proceed
14:06 Pulling over for a waypoint
14:18 Alert for car behind
14:47 Many pedestrians & crossing guard
15:06 DISENGAGEMENT: Slow to proceed with hand signals
15:38 Unable to re-engage (“Computer Issue”)
16:00 Remote assistance monitoring for construction
17:08 Stale status on vehicle operator’s phone
17:46 Infotainment draw distance
20:43 Light rail detection
21:01 Active emergency vehicle
22:30 Misc infotainment exploration
23:13 Arriving at final destination

Overall, I feel the car did handle the rain pretty well. At 7:35, I am not sure if the waymo needed to turn left. It seems it could have followed the white car. Maybe the waymo rerouted because of the "road closed - detour" sign? The waymo then reroutes again when it detects the construction zone and the worker moving cones. The waymo does handle snaking through the parking lot really well IMO.

There is the disengagement when the car was not going when the cross guard waved to go. Perhaps the car would have moved but was just too slow? We know waymo does have hand gesture recognition but clearly this is a data point they need to train on to improve the system.

Not sure what the computer issue was that prevented the car from re-engaging the autonomous driving but we see the path planning on the screen was still responding to things. So the autonomous driving seemed to still be working "in shadow mode".

I am hopeful that as waymo resolves these issues, Waymo will get even closer to being able to do reliable driverless in all conditions.
 
Last edited: