Personally, I think Waymo has the right approach. But they are realizing that actually solving FSD is harder than they thought. They underestimated the edge cases. And Waymo is fully committed to fully driverless which is harder. To deploy safe driverless robotaxis, you need to solve a lot more edge cases because the car has to be able to handle driving with no back-up. And really solving FSD is not just driving from A to B without hitting anything. Ideally, you also want the robotaxi to make smart decisions when interacting with other road users, know when to be assertive and when to be courteous to other drivers, drive smoothly, and mitigate accidents caused by others. Teaching a robot to do all that well is a very difficult problem.
Good question. Short answer is that I think Waymo was not really sure what business model to use at first and they underestimated the difficulty in actually solving FSD. Basically, they started as a FSD project for Google, got some good success with the technology but was not really sure how to convert that success into a real business. IMO, Krafcik was not a good CEO. He was too optimistic with the tech. He thought the FSD was ready for scaling when it wasn't ready yet. He was more flash over substance. In 2018, Krafcik apparently wanted to do a big multi-city marketing campaign to show off Waymo's FSD:
From the quote below, it seems Waymo also underestimated how hard that last 1% of FSD is. I think they kind of fell for their own hype when they were deploying robotaxis in Chandler and then had to pull back when they realized that the FSD was not solved yet. According to this, in 2017, they were planning to expand to 9 cities in just 18 months:
Source:
Bloomberg - Are you a robot?
I am more optimistic for the new co-CEO's, Dolgov and Mawakana. I think they have good complimentary skills for both the tech and the business side of things. They seem to have a better grasp of Waymo's business model (ride-hailing and trucking). They also appear to be more grounded and have a more realistic understanding of what it takes to actually solve FSD. I like that they have launched public testing in SF. I think it is a good sign that they are moving forward. I am also optimistic with the 5th Gen hardware, Simulation City and other ML work on prediction and planning that Waymo is doing. I feel that Dolgov and Mawakana will be about steady and concrete results.