Is Tesla closer than we think?

[shrineofchance]

Tesla told the CA DMV that that video was L3.

When/where did Tesla say that? These miles are reported as being in "autonomous mode" in Tesla's 2019 disengagements report to the California DMV.

In the recently released emails between Tesla and the California DMV, the terms "Level 3" and "L3" don't appear at all (although the term "Levels 3+" does).

What source are you referring to?

 

[shrineofchance]

In other states, each time there's a Waymo accident, the press would report it.

The press would have no way of knowing about every time a Waymo vehicle anywhere in any state contacted a stationary object.

 

[diplomat33]

So, in other words, Waymo's L4 is not technologically mature enough (or at least not demonstrably so) to be deployed to more than a tiny, token number of people — fewer than the engineers working on the L4.

Wrong. Waymo's tech has been deployed to thousands of people who ride in a fully driverless robotaxi every day in Chandler. The fact that Waymo's FSD is capable of driving consistently, every day, with no human in the car shows that it is very technologically mature. Is it perfect? no. But it is very mature FSD.

When/where did Tesla say that? These miles are reported as being in "autonomous mode" in Tesla's 2019 disengagements report to the California DMV.

In the recently released emails between Tesla and the California DMV, the terms "Level 3" and "L3" don't appear at all (although the term "Levels 3+" does).

What source are you referring to?

It was in an earlier batch of emails between Tesla and the CA DMV when the CA DMV asked about Autonomy Day.

 

[powertoold]

Waymo's tech has been deployed to thousands of people who ride in a fully driverless robotaxi every day in Chandler.

Do we have actual numbers for their public service?

I'd be surprised if they do more than a couple hundred rides a day.

 

[shrineofchance]

Wrong. Waymo's tech has been deployed to thousands of people who ride in a fully driverless robotaxi every day in Chandler. The fact that Waymo's FSD is capable of driving consistently, every day, with no human in the car shows that it is very technologically mature. Is it perfect? no. But it is very mature FSD.

If it's mature, why do they still have more engineers than vehicles? It's still very much an R&D project.

 

[diplomat33]

Do we have actual numbers for their public service?

I'd be surprised if they do more than a couple hundred rides a day.

Based on what I could find, Waymo did about 300 rides per day in 2019.

Chandler has a population of about 250,000. So Waymo One is available to thousands of people. And hundreds take rides every day. So my point is that Waymo is available to more than a tiny, token number of people as @shrineofchance said.

 

[diplomat33]

If it's mature, why do they still have more engineers than vehicles?

Are you trolling? Because FSD is hard and expensive. It takes time to solve FSD problems before you can deploy more vehicles safely. Just because you have mature FSD, does not mean that you can just deploy 1M robotaxis.

 

[shrineofchance]

300 rides per day is a tiny, token amount. This is a proof of concept, not a real consumer service.

 

[diplomat33]

If it's mature, why do they still have more engineers than vehicles? It's still very much an R&D project.

It is not an R&D project. It is a commercial robotaxi ride-hailing service that is available to the public in Chandler, AZ.

 

[shrineofchance]

Are you trolling? Because FSD is hard and expensive. It takes time to solve FSD problems before you can deploy more vehicles safely. Just because you have mature FSD, does not mean that you can just deploy 1M robotaxis.

Presumably, if you have one truly self-driving car, you can have unlimited of them by making identical copies of that car. Why does scaling up require R&D rather than simply manufacturing and logistics?

 

[diplomat33]

300 rides per day is a tiny, token amount. This is a proof of concept, not a real consumer service.

Waymo has deployed more robotaxis to the public than Tesla.

 

[shrineofchance]

There are only two possible conclusions:

1) Waymo has not yet truly solved FSD.

2) Waymo has solved FSD but they're not scaling up for some non-technological reason.

 

[diplomat33]

Presumably, if you have one truly self-driving car, you can have unlimited of them by making identical copies of that car. Why does scaling up require R&D rather than simply manufacturing and logistics?

There are only two possible conclusions:

1) Waymo has not yet truly solved FSD.

2) Waymo has solved FSD but they're not scaling up for some non-technological reason.

It's probably a bit of both.

Waymo has not completely "solved FSD" yet. But Waymo does have FSD that is good enough to be fully driverless in some conditions. Having true FSD that works in some areas does not mean you've solved all of FSD but it is a huge accomplishment.

And there are probably logistical problems to expanding as well.

Tesla has not solved FSD either.

 

[shrineofchance]

In Q2 2018 (three years ago), Waymo announced contracts with Fiat Chrysler and Jaguar to purchase up to 62,000 and 20,000 from each company, respectively:

Waymo to Buy Up to 62,000 Chrysler Minivans for Ride-Hailing Service (Published 2018)

Obviously, that never happened. Why not? My conclusion is that the technology wasn’t ready.

We don’t yet know why so many top executives at Waymo have left in the last 6 months: the CEO, CFO, Chief Safety Officer, head of manufacturing, head of automotive partnerships, head of operations strategy, head of IR, director of systems safety, and director of “Automotive Future”. My hunch is that this is a deliberate restructuring of the company, rather than a spontaneous development.

Maybe Sundar Pichai, the Alphabet board, or other higher-ups decided there needed to be a shake-up. In any case, it’s hard to see much, if any, progress toward commercialization of the technology. It still looks like a large R&D project that has fallen years behind on its commercial ambitions.

And this is because the technology hasn’t attained maturity as quickly as the former C-suite thought or hoped.

If Waymo had a truly viable consumer robotaxi product, they would rush to commercialize it at scale. But they started to rush to commercialize it at scale and then stopped. This indicates they ran into technical obstacles that remain unresolved.

 

[diplomat33]

Obviously, that never happened. Why not? My conclusion is that the technology wasn’t ready.

No, the Waymo FSD is probably not ready yet for safe, driverless, deployment in every single US city. Who said it was? But Waymo does have FSD that can do driverless in certain conditions. And, Waymo does have more robotaxis on public roads than Tesla does.

 

[diplomat33]

If Waymo had a truly viable consumer robotaxi product, they would rush to commercialize it at scale. But they started to rush to commercialize it at scale and then stopped. This indicates they ran into technical obstacles that remain unresolved.

Waymo has a viable consumer robotaxi product. They are already deployed in Chandler and will be deploying in San Franscisco soon (Waymo applied for a permit from the CA DMV). But yes, Waymo still has FSD problems left to solve.

Tesla does not any viable consumer robotaxi product.

 

[shrineofchance]

The Waymo vehicles without safety drivers rely on a few crutches:

• driving only at low speeds
• driving only along approved routes
• starting and stopping routes at approved spots
• falling back on remote assistance from humans
• using hand-annotated HD lidar maps that are updated daily
• constrained to a tiny geofenced area

Waymo’s ability to rely on driverless vehicles is as much or more a result of them making the problem smaller as making the solution bigger.

 

[shrineofchance]

Waymo has a viable consumer robotaxi product. They are already deployed in Chandler and will be deploying in San Franscisco soon (Waymo applied for a permit from the CA DMV). But yes, Waymo still has FSD problems left to solve.

Tesla does not any viable consumer robotaxi product.

It’s not commercially viable to pay 1,000 engineers to work on 100 vehicles that drive in a 50 square mile area.

 

[shrineofchance]

What is preventing Waymo from deploying 100,000 or 1 million robots is, rather than 100? It’s fundamentally the same problems of perception, prediction, and planning that everyone is struggling with.

There is no guarantee that Waymo will solve the technical challenges to get to the point where they can deploy 100,000 robotaxis. This is an ongoing research and engineering challenge.

Tesla is working on the same fundamental technical challenges in perception, prediction, and planning, but on a different problem than Waymo in terms of the application:

• driving at any legal speed
• driving along any route
• starting and stopping anywhere a vehicle can park
• falling back on an attentive human driver
• using real time sensing rather than HD maps
• minimal geofencing

This is a much bigger problem than the problem Waymo has carved out for themselves in Chandler. It requires a much bigger solution with fundamentally better AI. It’s an apples-to-oranges comparison.

 

[Tam]

The Waymo vehicles without safety drivers rely on a few crutches:

• driving only at low speeds

Not true. Chandler does have freeways and though not very often, its driverless without a safety driver does take highways at times.

• driving only along approved routes

True. That's why it's not L5

• starting and stopping routes at approved spots

Ideally, it should be just like train- or bus-approved unchangeable stopping spots to simply the task but Waymo seems not to do that right now and because of random, unplanned stop/pickup locations within the 50 square miles, the car might miss the exact spots at times.

• falling back on remote assistance from humans

True. But remote operator does not "drive" the car remotely. No controls for pedals remotely. The remote operator inputs additional information as if the human is an additional sensor for the system to make a decision: pause, forward, reverse...

• using hand-annotated HD lidar maps that are updated daily

I am not familiar with that but ideally, the map should be updated as frequently as possible (the traffic cones might not be there an hour ago but now they are...)

• constrained to a tiny geofenced area

True. No argument there

Waymo’s ability to rely on driverless vehicles is as much or more a result of them making the problem smaller as making the solution bigger.

It has pretty much solved the problem of avoiding crashes.

It has not solved the problem of intelligence. When it doesn't know what to do, it pauses which means it doesn't crash at all during that state but that is not intelligent enough to figure out what to do next to get out of the pause.

If it's mature, why do they still have more engineers than vehicles? It's still very much an R&D project.

If Waymo had a truly viable consumer robotaxi product, they would rush to commercialize it at scale. But they started to rush to commercialize it at scale and then stopped. This indicates they ran into technical obstacles that remain unresolved.

Again, Waymo is mature at avoiding crashes and they thought that would be the end of the problems so they could scale up. However, after a while, they realize that the system is not intelligent enough to figure out how to get out of a situation that made it pause.

Waymo's intelligence to make decisions that were not inputted previously is still a very early phase.

I agree that its product is not viable right now, not because of lacking driving skills to avoid crashes but because its intelligence to make decisions are still very infantile.