If this is true, then why aren't these companies testing their generalized solutions in more locations at once?
Over 10 years, Waymo has tested in 25+ cities across the US. And from public info we have, Waymo is doing autonomous driving in SF, LA, Phoenix, Orlando and Austin at once. Now, if you are asking why they don't test in even more locations at once, I think the answer is that they don't need to. It would not give them any significant benefit. Waymo has lots of data already from the places that they are testing and when Waymo needs specific data, they simply go and get the data they need. For example, when they needed dense urban data, they went to NYC, when they needed rain data, they went to Bellevue, WA State and in Orlando, FL, when they needed snow data, they went to Michigan.
But I am not making up that Waymo is generalized. Waymo co-CEO said so:
It all allows the
#WaymoDriver to generalize across cities, which enabled us to scale in SF and PHX simultaneously this year, and showed incredible performance in LA right from the get-go.
Source:
So, the proof that Waymo is generalizing well is the fact that they were able to scale in PH and SF simultaneously and the fact that they launched the same Waymo Driver in LA and he says it performed incredibly well from the get-go. This implies that they did not need to do any overfitting to get it to work in LA. So I think this tweet directly refutes your claim that they need to overfit city by city.
What benefit do they have to assigning a fleet of 50 to a single city, when they could deploy one vehicle to 50 of the largest cities in the US to demonstrate their technology?
The benefit in assigning 50 to a single city is that they can launch a ride-hailing service in that city. Waymo's whole business model is doing safe, reliable driverless robotaxis. Deploying 1 car in 50 cities would not speed up their research or speed up their scaling of a robotaxi service and it would take resources away from actually deploying a robotaxi service. It would basically just be a stunt. It would not help them in their overall mission.
Just so happened to be looking at how long it's been taking Waymo to begin a new service in NYC, and found a familiar face asking a question:
They began that mapping work in 2021. It's 2023 now, and we still don't have insight on when they'll begin service. Surely a generalized solution doesn't take that long to expand.
This is not proof of poor generalization because Waymo was not planning to launch robotaxis in NYC. They were only in NYC to collect data, not to launch a new service. In fact, NY does not allow driverless so Waymo cannot start a robotaxi service in NYC even if they wanted to.
Just because Waymo tests in a city does not mean they plan to launch a ride-hailing service. Many times, Waymo tests in a city, just for data and validation. Unless Waymo actually announces a ride-hailing service like they did with LA, we should not assume that they planning to launch a ride-hailing service in that city.