The autonomous-vehicle startup Zoox, which was founded in 2014, decided to focus on one of the hardest testing environments first. Rather than starting in a quiet suburb, the company chose to begin testing its technology in San Francisco.
While warm weather and the absence of snow make it easier for self-driving vehicles to operate; dense, urban environments are more difficult to navigate than slower-paced subdivisions that normally boast wide streets that are typically clear of most obstructions.
I took a ride in one of Zoox's test vehicles earlier this week as it drove through one of the company's "challenge" routes. The route featured steep roads, narrow streets with parked cars on either side, and an unprotected left turn at a six-way intersection.
While the ride wasn't perfect, the vehicle performed as well as the average human driver, and the ride was more comfortable than the Uber and Lyft I took to and from the airport during my trip to San Francisco.
Here's what it was like.
The ride began outside one of Zoox's offices in San Francisco.
The ride took about 30 minutes.
With a few exceptions, the vehicle drove as well as the average human driver.
It even handled Lombard Street, famous for its series of sharp, steep turns.
If I hadn't known I was in an autonomous vehicle, I would never have thought that I was being driven by a computer.
The vehicle followed cars in front of it at a safe, but not-too-lengthy distance; correctly read and tended to come to a smooth stop at red lights and stop signs; confidently navigated narrow roads with cars parked on both sides; and had an impressive understanding of when pedestrians would or would not move in front of it.
And it was able to complete an unprotected left turn at a six-way intersection.
It was a little more cautious than I would have been, but I'd rather an autonomous vehicle be too careful than too aggressive.
There were a few hiccups, though.
On a few occasions, the vehicle stopped abruptly without having a clear reason to do so, and it didn't always seem comfortable driving through a four-way intersection when a vehicle was driving toward it in the other lane. My vehicle's safety driver appeared to take control from the autonomous-driving software at least three times.
In an interview after my ride, Zoox CTO and cofounder Jesse Levinson told me that the company's vehicles are better than human drivers at following the rules of the road, but worse at handling unusual events.
So while Zoox vehicles are better than humans at stopping at stop signs and following speed limits, Levinson said, they're worse at predicting the behavior of vehicles being driven erratically, because humans, of course, can be unpredictable.
My ride in Zoox's vehicle was more comfortable than those I took through Uber and Lyft on my way to and from the airport during my trip to San Francisco.
The drivers for my Uber and Lyft rides tended to stop and start more forcefully than the Zoox vehicle.
It's important to note that the route used for my Zoox ride has likely been tested extensively, which means it's difficult to draw many conclusions about Zoox's autonomous-driving technology beyond my specific and relatively brief experience.
But overall, I was impressed.
This was the third ride I'd taken in an autonomous vehicle. The first two came with Yandex and Aptiv in Las Vegas in January.
Both of those rides were also impressive, but the routes they took were much less challenging. My ride with Zoox showed that its technology, at least in the narrow confines of a company-selected route, can handle at least some of the challenges of city driving.