I Get Back In A Waymo To Ride In San Francisco With A Top Waymo Developer And It’s Good (+Video)
I recently took a ride in a fully autonomous Waymo vehicle in San Francisco. It was my first ride in many years — I had been a member of the early team while it was part of Google. My guide on the ride was Andrew Chatham, whom I had worked with back then. He is now a Distinguished Engineer, managing fleet logistics and many other things, and reporting directly to Waymo’s co-CEO.
(The video above includes my discussions with Chatham, with information not necessarily found in this text article.)
In that earlier era, vehicles of course needed a safety driver — I did that myself on rare occasions — but this day the driver seat would stay vacant and we would drive a much more complex environment. Many others are posting videos and reporting the same thing — the rides are smooth and confident, with nothing more than minor nits to comment on. Errors were frequent enough a decade ago, similar to Tesla FSD, and the road has been long. Our vehicle wandered around the Cole Valley and Haight-Ashbury areas of San Francisco in the daytime. Traffic and pedestrians were common but not congested in the early afternoon. We encountered and pulled aside for an emergency vehicle, handled many pedestrians and double-parked vehicles, and made a few virtual stops. (My main complaint was that some of the stops stuck out into traffic a bit, a complaint that many robotaxis have received. It’s not uncommon for taxi drivers, though.)
The problem, though, is that one smooth ride, or even 100 smooth rides, don’t tell you what you want to know. They tell you the team is in the race, but the real bar is to do tens of thousands of smooth rides without incident, and no single person’s experience or video can convey that.
A video of part of my ride is included, where you can see the vehicle handle pedestrians, double parked cars, unprotected left turns, speed bumps, cyclists and many other things from a city, plus hear my discussion and questions for Waymo. Of course, Chatham had to dodge questions about things they have not decided to disclose. I’ve also edited out some off-topic or personal discussions and sped up those sections.
The strongest message conveying that was Waymo’s willingness to give me, and many members of the public, and a senior executive, that ride — with nobody up front. That tells us that they did a lot of analysis and had intense meetings where they looked at their data and concluded it was safe. An incident with anybody would set their project back and they now are ready to face that challenge. To be fair, an even better test, which they are letting members of the public do, is to let the passenger choose the time and route. I’ve had a number of press demo rides in various cars that went fine, but they picked the route and surely had tested it several times before.
Chatham agreed that a competent team can make a demo video in “a couple of quarters.” That’s not enough. They are proud of where they are, and particularly that they drive all day with the public while others might drive only at night.
Waymo just expanded autonomous operations in SF to include the most complex downtown area, however they are still limiting those rides only to employees. While this shows they think the risk of a safety incident is very low, they are not ready to take that risk with the public, and more to the point let the public see the non-safety-related quirks and rough edges of a new territory. They also announced they had gotten their permit to operate in Los Angeles. LA driving isn’t as hard as SF, but it’s a much larger city, so if they want a decent sized service area with tolerable wait times, that will need a big fleet. Waymo currently has 700 vehicles in all their cities.
Before long, my ride might be this custom design made by Chinese maker Zeekr/Geely for Waymo. It ... [+]WAYMO I like to ask every developer what are the blocking factors in the way of going into production. I have an article and video series on answers to that, but Chatham, who has been managing the software for fleet operations says the difficulty of that should not be underestimated. Aside from dispatching and directing vehicles and bringing in the remote operators, vehicles must be predictively positioned to reduce wait times, and also be charged, since Waymo now uses electric vehicles in all new deployments, including SF. You need systems to deal with the things that go wrong, according to Chatham, in some cases having teams to retrieve a vehicle or get it out of stuck spot when remote operations can’t solve the problem. “Not everybody handles that well,” according to Chatham.
Tesla FSD driving to the ride
The ride confirms, to the limited extent any ride can do that, what is commonly said — Waymo is the clear leader. While Cruise has definitely made strides this year, and does operate with employees in the day, Chatham is correct in pointing out that driving during the day is a big step up from operating late at night. It’s harder to judge the Chinese companies who are operating in the daytime, without going to China.
It’s amusing to see those who tout Tesla as the leader because their system fails in every area rather than just failing in a limited set of areas. I drove to this ride using Tesla FSD, and as always with any significant ride, an intervention was needed. In this case, one was needed just as I approached the Waymo pickup point as the car imagined a one-way road and drove on the wrong side in the same territory the Waymo would handle without issue. When the standard is to do tens of thousands of rides without such problems, a system that has trouble doing one ride simply isn’t in the running, no matter how many roads it is willing to attempt.
What’s less clear is whether Andrew Chatham’s statement that operational issues are one of the big remaining blockers is correct. While there is definitely work to be done to make that smooth, nobody thinks that’s intractable and it can be solved with time and money with little uncertainty, though making it cheap needs more effort. Scaling up a robotaxi fleet is very far from trivial, but you don’t want to do it until you are sure you are ready in all other ways — but once you are ready, you can do it.
Soon, I’ll take a ride in a Cruise vehicle to track their progress. As you may have seen on the AutonoMap released yesterday, services areas are popping up around the world and the game is on.
Tue, 2023-02-14 12:22
Apple awarded patent on Autonomous Vehicle Lane Change
Add new comment