Dmitri Dolgov, who recently took over as Waymo’s co-chief executive after the departure of John Krafcik, an automobile industry veteran, said the company considers its Arizona service a test case. Based on what it has learned in Arizona, he said, Waymo is building a new version of its self-driving technology that it will eventually deploy in other geographies and other kinds of vehicles, including long-haul trucks.

The suburbs of Phoenix are particularly well suited to driverless cars. Streets are wide, pedestrians are few and there is almost no rain or snow. Waymo supports its autonomous vehicles with remote technicians and roadside assistance crews who can help get cars out of a tight spot, either via the internet or in person.

“Autonomous vehicles can be deployed today, in certain situations,” said Elliot Katz, a former lawyer who counseled many of the big autonomous vehicle companies before launching a start-up, Phantom Auto, that provides software for remotely assisting and operating self-driving vehicles when they get stuck in difficult positions. “But you still need a human in the loop.”

Self-driving tech is not yet nimble enough to reliably handle the variety of situations human drivers encounter each day. They can usually handle suburban Phoenix, but they can’t duplicate the human chutzpah needed for merging into the Lincoln Tunnel in New York or dashing for an offramp on Highway 101 in Los Angeles.

“You have to peel back every layer before you can see the next layer” of challenges for the technology, said Nathaniel Fairfield, a Waymo software engineer who has worked on the project since 2009, in describing some of the distractions faced by the cars. “Your car has to be pretty good at driving before you can really get it into the situations where it handles the next most challenging thing.”

Like Waymo, Aurora is now developing autonomous trucks as well as passenger vehicles. No company has deployed trucks without safety drivers behind the wheel, but Mr. Urmson and others argue that autonomous trucks will make it to market faster than anything designed to transport regular consumers.

Long-haul trucking does not involve passengers who might not be forgiving of twitchy brakes. The routes are also simpler. Once you master one stretch of highway, Mr. Urmson said, it is easier to master another. But even driving down a long, relatively straight highway is extraordinarily difficult. Delivering dinner orders across a small neighborhood is an even greater challenge.

“This is one of the biggest technical challenges of our generation,” said Dave Ferguson, another early engineer on the Google team who is now president of Nuro, a company focused on delivering groceries, pizzas and other goods.

Mr. Ferguson said that many thought self-driving technology would improve like an internet service or a smartphone app. But robotics is a lot more challenging. It was wrong to claim anything else.

“If you look at almost every industry that is trying to solve really really difficult technical challenges, the folks that tend to be involved are a little bit crazy and little bit optimistic,” he said. “You need to have that optimism to get up everyday and bang your head against the wall to try to solve a problem that has never been solved, and it’s not guaranteed that it ever will be solved.”

Uber and Lyft aren’t entirely giving up on driverless cars. Even though it may not help the bottom line for a long time, they still want to deploy autonomous vehicles by partnering with the companies that are still working on the technology. Lyft now says autonomous rides could arrive by 2023.

“These cars will be able to operate on a limited set of streets under a limited set of weather conditions at certain speeds,” said Jody Kelman, the executive of Lyft. “We will very safely be able to deploy these cars, but they won’t be able to go that many places.”