Waymo To Begin No-Driver Operations with Public In SF; Zoox Accident

The California PUC granted Waymo a permit to operate 24 hours/day in San Francisco taking select members of the public for rides with no safety driver in the vehicle. Waymo says it will begin this shortly. This comes on the heels of them expanding such service in Phoenix, as reported in my article on how the death of self-driving cars has been greatly exaggerated earlier this week.

This service will be with “trusted testers” rather than members of the broad public that can ride in Chandler and Phoenix, Arizona. Limiting ridership can be useful when demand is higher than supply, and also to learn from people who ride it repeatedly, but it has another less noble purpose, namely that riders can be under NDA and not exposing any problems to the public. (Waymo previously has required this of “trusted testers” but says they will very quickly transition to not needing an NDA.) Once a service allows members of the general public to ride with no conditions, it’s a declaration that “we’re confident we are not going to do anything embarrassing.” Cruise, which is Waymo’s rival in SF, has been serving the public for some time, and on Nov 16 announced they will be doing daytime rides with GM employees. Cruise had previously only been operating with passengers after 10pm on much calmer streets.

Waymo also announced it will deploy in Los Angeles soon (it has been testing with staff there.) In LA they also revealed their new robotaxi model, built by Geely under the Zeekr brand in China. I will have more comment when I get to see it, though I am disappointed the front seats don’t swivel to allow more social settings when a group of more than 2 go for a ride. (Facing backwards is less comfortable for some, but also safer in a forward crash.)

Every project seems to go through a series of milestones which demonstrate how confident they are in the system:

1. Safety driver, no passengers (pax): Not that much confidence
2. Safety driver, external pax: Safety drivers are pretty good so not too much risk, but a bit more. The big step is you are letting the public see operations in a non-cherrypicked way.
3. No safety driver, vacant: Big step, but no passenger risk and nobody watching inside.
4. No safety driver, employee: Big step, but mistakes don’t go public unless there is a serious injury.
5. No safety driver, NDA public: Medium step
6. No safety driver, general public: You are exposing all your foibles and must think you’re pretty good.

Each of these steps also can vary based on what network of roads and driving conditions the vehicle handles (known as the O.D.D.) including Cruise’s option of night-driving vs. daytime. Teams have also made an inflated deal about their permits which allow them to charge money for rides — nobody is charging money to get revenue, this is more simply the ability to experiment with how people respond to having to pay, and a way to limit demand.

Limiting demand is a real thing when you have a small fleet. The public are being let in to learn how to serve the public, and if it’s very hard to get a ride, you only learn what single rides are like and how people react. That’s decently understood, now. To plan a real service you have to study people who come to depend on using the service, who possibly consider giving up one of their cars because of it. To see that, you need a service that can offer a ride any time, with not that long of a wait.

Non-news from Zoox

Alert eyes noticed a Zoox test car which rear ended a truck in Las Vegas. It’s non-news because Zoox reports the vehicle was not in self-driving mode. It appears, however, that the safety driver got confused over what mode it was in, and thus didn’t hit the brakes because the safety driver expected the car to handle that.

This is an issue that people worry about, called “automation complacency.” That safety driver got so used to autonomous driving that they didn’t pay attention to whether the car was in automous mode very well, and didn’t react in time when it was clear it wasn’t. Generally, safety drivers do very well. In fact, even amateur Tesla owners with Tesla’s FSD pre-alpha prototype seem to take over capably and reports of accidents with that system are surprisingly rare. In part, they do that because of the very low quality of the Tesla FSD system — it’s hard to get complacent in such a situation, and people are paying attention. As a system gets better, it gets easier for even a paid and trained safety driver to make a mistake, it seems. (Though this is nothing like the Uber safety driver who decided to completely ignore all the rules and watch TV, resulting in a tragic fatality.)

Aurora

Aurora contacted me to contest how I wrote earlier this week in my article about earlier stage efforts that are more tractable than robotaxis that they had pivoted to trucking. Aurora was always interested in all vehicles, including robotaxis and trucking, but almost all their announced progress and all events on their roadmap have related to trucking. They did announce that they had built a prototype robotaxi with Toyota for their Aurora Connect service, but at this point, lots of people have showed off a prototype and driven around with insiders — it now takes much more than that to show a serious robotaxi effort.