Waymo's self-driving Jaguar I-Pace begins testing on public roads

[electronblue]

Yes and that goes along with my point: people only see Waymo driving around in a small 80 square mile area so they may think that is all the cars are capable of. And removing the safety driver permanently would be a big deal but I think some people might ask the question: was Waymo able to remove the safety driver only because the cars repeated so many drives in that same small 80 square mile area that they got really good on those drives or because the cars are really good enough that they don't need safety drivers anywhere?

But looking at one location only (where they sell a service) would be silly given that Waymo operates test fleets in a dozen locations and publishes tons of mileage data from this. Personally I’m confident, if they wanted to, Waymo could autonomously drive within the entire U.S. with superior results to what Tesla can do today. Not perfect perhaps but superior to what Tesla can do today.

The more interesting question of course is the trajectory of things. As Waymo’s system matures and Tesla builds their own, where will these trajectories head and meet at. Tesla has some benefits we’ve discussed but equally Waymo and others have other benefits — and I think it is pretty well established the competition of Tesla has the lead so far.

 

[Daniel in SD]

Yes and that goes along with my point: people only see Waymo driving around in a small 80 square mile area so they may think that is all the cars are capable of. And removing the safety driver permanently would be a big deal but I think some people might ask the question: was Waymo able to remove the safety driver only because the cars repeated so many drives in that same small 80 square mile area that they got really good on those drives or because the cars are really good enough that they don't need safety drivers anywhere?

I don't think people realize how much more difficult it is to do 100,000 five mile drives in a city than to do a single cross country drive.

 

[diplomat33]

The more interesting question of course is the trajectory of things. As Waymo’s system matures and Tesla builds their own, where will these trajectories head and meet at. Tesla has some benefits we’ve discussed but equally Waymo and others have other benefits — and I think it is pretty well established the competition of Tesla has the lead so far.

As I see it, there are 2 dimensions: fleet size and FSD quality. Tesla is high on fleet size and low on FSD quality. Waymo is low on fleet size but high on FSD quality.

I don't think people realize how much more difficult it is to do 100,000 five mile drives in a city than to do a single cross country drive.

Relating it to Tesla, a better question would be: Which is easier, 1 Waymo car doing the same 5 mile city drive 100,000 times or a 100,000 Teslas doing one 5 mile city drive? I say this because Tesla's advantage is not doing a cross country FSD drive. Tesla's advantage is that they can do 500,000 miles of FSD testing by just having 100,000 cars do 5 miles of self-drive.

 

[electronblue]

As I see it, there are 2 dimensions: fleet size and FSD quality. Tesla is high on fleet size and low on FSD quality. Waymo is low on fleet size but high on FSD quality.

Yes.

However there is a third dimension — and that is how do those fleet sizes and FSD qualities (and other qualities) correlate with future progress. A simplification to illustrate the point: If you believe fleet size is the key to high FSD quality, you can expect Tesla to do better. If you believe fleet size to be irrelevant (or secondary) to reaching high FSD quality, you might believe Waymo will remain ahead.

Finally, things also change if you believe the fleets and/or FSD of either of these players could become irrelevant due to, say, them sporting a sensor suite that ends up being insufficient.

 

[Daniel in SD]

As I see it, there are 2 dimensions: fleet size and FSD quality. Tesla is high on fleet size and low on FSD quality. Waymo is low on fleet size but high on FSD quality.

I'm of the opinion that FSD quality is all that matters until companies actually get it working. I suspect, if they thought it would help, Cruise and Waymo could easily have much larger fleets. They have so much money to play with.

Relating it to Tesla, a better question would be: Which is easier, 1 Waymo car doing the same 5 mile city drive 100,000 times or a 100,000 Teslas doing one 5 mile city drive? I say this because Tesla's advantage is not doing a cross country FSD drive. Tesla's advantage is that they can do 500,000 miles of FSD testing by just having 100,000 cars do 5 miles of self-drive.

Those would be equally hard. If Tesla can do 100,000 random five mile drives without a single accident, with the current sensor suite, then they're a trillion dollar company.

 

[diplomat33]

I'm of the opinion that FSD quality is all that matters until companies actually get it working. I suspect, if they thought it would help, Cruise and Waymo could easily have much larger fleets. They have so much money to play with.

You are right that these companies could easily pump out a lot of cars if they wanted to. So why aren't they? I think part of of their strategy is guided by the fact that they don't have OTA update capability so they can't update the fleet on the go. Putting out cars with their current state of FSD would just mean a lot of cars with "bad" FSD out there. So it is better for them to wait until they finish FSD and then put the perfect FSD on new production cars. Tesla has OTA update capability so they are fine with putting out "unfinished" FSD because they can just update the fleet on the go when they have better software.

Those would be equally hard. If Tesla can do 100,000 random five mile drives without a single accident, with the current sensor suite, then they're a trillion dollar company.

I think that is the whole idea behind Tesla's strategy. When Tesla gets to "feature complete", they can use the fleet to do 100,000 random five miles on the FSD software, get feedback, refine the software, push it out, do another 100,000 random 5 fives miles of FSD, get feedback, rinse and repeat until they get down to 0 disengagements. Tesla hopes that this strategy will get them to 0 disengagements and when it does, like you said, ka ching, $$$$$$$$$$$$$$.

 

[electronblue]

@diplomat33 I’m pretty sure Waymo and Cruise have OTA capability — or at least that it has got nothing to do with how big a pilot/test/development fleet they run.

I mean, they can drive a Waymo remotely. Of course they can update the software too. But even if they couldn’t, I seriously doubt that would matter at this stage at all.

I believe the fleet size for the likes of Waymo and Uber is more a function of what is needed for development and testing. These companies don’t think they need a massive fleet for that and indeed even Tesla collects a miniscule amount of their ”miles” too.

As for Tesla’s approach, I do agree Tesla’s approach is different (and their hopes are probably quite close to what you described) — so just trying to describe what I think is Waymo’s and Cruise’s approach to this.

 

[Tam]

...Cruise...

Now that you mentioned Cruise Automation of GM, there's new sensational dirt:

https://jalopnik.com/gms-cruise-self-driving-prototypes-are-riddled-with-tec-1835414849

I guess it has not worked out a redundancy system nor the shutting down procedure when there's a computer crash because the system stopped working during a demo to Honda ($2.75 investor) and the car was still keeping on rolling then the safety driver took over at that time.

 

[Krugerrand]

I think that car should go to the casting call for the next Ghostbusters movie.

 

[Daniel in SD]

Now that you mentioned Cruise Automation of GM, there's new sensational dirt:

I thought that article was funny. It says they have a close call every 450 miles. If people think that's bad wait until Tesla releases "automatic city driving". The videos from that will be a gold mine for clickbait sites.
Keep in mind that they're driving around San Francisco 24/7. It's a very challenging driving environment. I do wonder how they know when to take over since I'd expect that I'd have a "close call" rate higher than once every 450 miles driving in SF.
Here's an article on how Cruise trains their test drivers:
How GM trains human drivers to monitor its autonomous cars
I think Tesla may run into the same problems and be forced to have a training program for people using beta FSD.

 

[AlanSubie4Life]

Tesla may run into the same problems and be forced to have a training program for people using beta FSD

This has to be your favorite bromide. My favorite bromide is that FSD is level 2, so there will be no need for a training program.

 

[Daniel in SD]

This has to be your favorite bromide. My favorite bromide is that FSD is level 2, so there will be no need for a training program.

But how can you tell that Cruise vehicles are not level 2?
They should just tell their test drivers that they are level 2 and then there would be no safety issues and no need to train them.

 

[AlanSubie4Life]

But how can you tell that Cruise vehicles are not level 2?
They should just tell their test drivers that they are level 2 and then there would be no safety issues and no need to train them.

There you go again. There are so many ways to skin this cat. As I said, this is definitely your favorite talking point. In the end, this is a philosophical, existential discussion, not a logical one. If Level 3+ does not exist, do you need a permit and/or training? If not, how will Level 3+ ever exist? At least that is the way things are currently treated by the California DMV. That may change of course. Which I guess is your point.

 

[diplomat33]

@diplomat33 I’m pretty sure Waymo and Cruise have OTA capability — or at least that it has got nothing to do with how big a pilot/test/development fleet they run.

The test cars probably have OTA update capability. I was thinking of the commercial GM cars that the public can buy. They don't have OTA capability. Again, it goes to the difference in strategy. Tesla is doing their FSD updates on commercially available cars rather than just test cars.

Now that you mentioned Cruise Automation of GM, there's new sensational dirt:

https://jalopnik.com/gms-cruise-self-driving-prototypes-are-riddled-with-tec-1835414849

I guess it has not worked out a redundancy system nor the shutting down procedure when there's a computer crash because the system stopped working during a demo to Honda ($2.75 investor) and the car was still keeping on rolling then the safety driver took over at that time.

That article sure makes Cruise look bad. Yikes. Just goes to show how one article can make a FSD program look really good or really bad. Cruise had that announcement about doing a record number of unprotected left turns that made it look like their FSD was really good but this article shows that the program still has some serious issues. It also explains why the CEO of GM has backtracked on her previous promise of removing the steering wheel and pedals by the end of this year.

I thought that article was funny. It says they have a close call every 450 miles. If people think that's bad wait until Tesla releases "automatic city driving". The videos from that will be a gold mine for clickbait sites.

That's why Tesla is going to wait until they are confident that "automatic city driving" is good enough for the public. Tesla is not dumb. They are not going to release a feature before it is ready and cause a ton of accidents. Heck, we know that "automatic city driving" is in shadow mode on our cars now because someone hacked their car and enabled it but Tesla has not released it to the public because it is not ready yet.

 

[Daniel in SD]

There you go again. There are so many ways to skin this cat. As I said, this is definitely your favorite talking point. In the end, this is a philosophical, existential discussion, not a logical one. If Level 3+ does not exist, do you need a permit and/or training? If not, how will Level 3+ ever exist? At least that is the way things are currently treated by the California DMV. That may change of course. Which I guess is your point.

How do you explain the DMV’s decision about Uber’s “level 2” vehicles?

 

[Daniel in SD]

The test cars probably have OTA update capability. I was thinking of the commercial GM cars that the public can buy. They don't have OTA capability. Again, it goes to the difference in strategy. Tesla is doing their FSD updates on commercially available cars rather than just test cars.

The 2020 Cadillac CT5 will have OTA updates and GM claims all their car will have the capability by 2023. They’re not claiming that any of their vehicles will have FSD capability though.

 

[electronblue]

The test cars probably have OTA update capability. I was thinking of the commercial GM cars that the public can buy. They don't have OTA capability. Again, it goes to the difference in strategy. Tesla is doing their FSD updates on commercially available cars rather than just test cars.

I guess I don’t always follow the same logic as you. I don’t see how commercial cars would be relevant as neither Waymo or Cruise sell any cars to the public. All they have are their internal fleets. And the reason they don’t sell them elsewhere is they are not finished with the technology yet so in their strategy there is no point.

I acknowledge Tesla has a different strategy with some deployment and testing advantages as well as the downside that they have made a bet on a rather more limited sensor suite and are ”stuck with it” because deploying something Waymo/Cruise like would not have been realistic so early on.

Anyway, my point is I don’t think Waymo/Cruise fleet sizes are dictated by either limitations in building them or anything like OTA capability, but simply their perceived need for the fleet — which is limited in number. Their self-driving technology is not ready yet and in their plan they don’t need an over-sized test fleet at this stage.

It is of course perfectly okay to disagree with their plan but I think that is their plan.

 

[electronblue]

That article sure makes Cruise look bad. Yikes. Just goes to show how one article can make a FSD program look really good or really bad. Cruise had that announcement about doing a record number of unprotected left turns that made it look like their FSD was really good but this article shows that the program still has some serious issues. It also explains why the CEO of GM has backtracked on her previous promise of removing the steering wheel and pedals by the end of this year.

That's why Tesla is going to wait until they are confident that "automatic city driving" is good enough for the public. Tesla is not dumb. They are not going to release a feature before it is ready and cause a ton of accidents. Heck, we know that "automatic city driving" is in shadow mode on our cars now because someone hacked their car and enabled it but Tesla has not released it to the public because it is not ready yet.

You do realize that in the space of the same message you made Cruise look bad for testing their system and taking their time to do so, while somehow simultaneously making Tesla look good for simply taking their time and not even showing us any testing results!

I sense a double standard. But then this has been the way here in recent times. Tesla is judged by the loftiness of their promise and everyone else is judged by the testing results seen.

I guess it is understandable, because were we to judge Tesla by their test results seen (and in many cases lack thereof), the only possible conclusion would be utter condemnation.

 

[Phrixotrichus]

That's why Tesla is going to wait until they are confident that "automatic city driving" is good enough for the public. Tesla is not dumb. They are not going to release a feature before it is ready

wait...WHAT!?

Tesla has been doing nothing else but releasing features that are basically beta versions, that`s especially true for anything AP but also for features like the infamous NN-powered wipers.
Heck for a long time AP didn`t even require any driver feedback and people were doing videos sitting in the backseat and similar stuff, and the feature didn`t even recognize crossing semis at that point...
It´s exactly because Tesla is doing this Beta-Release stuff that the OTA capability is such a major point in their strategy.
Every single driver is a beta tester.