San Francisco's North Beach neighborhood is how is eroticism different from euphorianotoriously tricky for any driver. Narrow roads, gobs of tourists, steep inclines, double-parked delivery vans, never-ending construction, and maddening six-way intersections run rampant there. Take away the human driver, and you've got a potential recipe for disaster.

Nevertheless, that's what Zoox, a self-driving startup, is attempting. But instead of going to the sunny, wide boulevards of Arizona or Silicon Valley, it's focusing on navigating through the windy roads of a dense, packed urban area. It's going to take a while to really get there, especially with Zoox's vision of a robo-taxi service in a made-for-self-driving vehicle that's still in development.

On a recent Zoox demo ride through North Beach and other crowded San Francisco neighborhoods, a safety driver was practically glued to the steering wheel and kept his eyes focused intensely on the road. Another safety engineer monitoring the self-driving computer system sat in the passenger seat. The humans never took over, but it was a nerve-wracking drive.

When we came upon a double-parked van on a narrow road, instead of zipping around it like anyone would normally do, we waited for a "safe" moment to go around. We waited so long the double-parked driver eventually moved along. Problem solved? Not really. In a 20-minute demo ride we encountered a constant stream of "edge" cases that other demos in Las Vegas and Phoenix would freak out about. Driving in San Francisco is never a typical experience.

For Zoox director of computer vision Sarah Tariq, the chaos of testing in a place like San Francisco is the point. "We want to test in the hardest places," she said after our ride in a modified Toyota Highlander. "Whatever we learn from here, it feeds into our launch." If the Zoox cars can't handle pedestrians literally blocking the entire intersection at the famous crooked Lombard Street, they're not ready for other "easy" driving scenarios. For the record, the Zoox test car politely scooted closer to the pedestrians until they realized they were blocking the road and moved.

Cruise is also attempting to start an SF-wide robo-taxi service, but that's already hit snags and delays. Just this week Cruise's parent company presented a new vehicle meant for ride-share without a driver's seat or steering wheel. The Cruise Origin six-seater is supposed to supplement Cruise's eventual taxi service in a fleet of Chevy Bolts. Honda and GM are collaborating on what it considers the first production-ready autonomous vehicle, and the companies emphasized it's not a concept. But nothing's here yet.

Compare all this to a Waymo driverless ride I took the week before in Chandler, Arizona. I took a very real and very empty Chrysler Pacifica minivan for a roundtrip journey from a hotel parking lot to a shopping mall parking lot. Except for one stressful moment at a flashing yellow left turn signal, the Waymo handled the trip just as it was programmed. I'd even say the Waymo drove better than most people I know. It made the perfect swooping turn into the parking lot. 10/10.

This was through the Waymo One app-based service available for a fee like a Lyft or Uber ride for riders who live in the area. No demos, but the real thing.

This Tweet is currently unavailable. It might be loading or has been removed.

Autonomous trucking is another space within self-driving that's moving along smoothly. It's not just the wide, open, predictable suburban roads that are compatible with real progress. The open highway is a great place for self-driving systems to take over. On stretches in the middle of the country, like Texas and New Mexico, this is especially true. This week, Waymo is testing its autonomous semis in those areas.

This Tweet is currently unavailable. It might be loading or has been removed.

Cruise's co-founder Kyle Vogt recently wrote a blog post discouraging the very comparisons I just made based on one-off demo rides. I see how this is clearly an apples-to-oranges comparison with different regulatory environments. As Vogt wrote, "Keep in mind that driving on a well-marked highway or wide, suburban roads is not the same as driving in a chaotic urban environment. The difference in skill required is just like skiing on green slopes vs. double black diamonds."

It's well known that driverless miles aren't all equal. Autonomous vehicles will always be different in a busy city than in car-centric sprawl. But no matter, places like suburban Chandler are off and running with the first generations of true AV rides while places like San Francisco are still in hyper-supervised testing mode. That's not going to change anytime soon.