In the world of self-driving, the strong leaders are now Google/Alphabet’s Waymo, and GM’s Cruise. Both are headquartered in the San Francisco Bay Area, and both have thus been keen to deploy in the city of San Francisco itself. No wonder, it’s their hometown. Driving it is a challenge but an important one to handle. On the other hand, it doesn’t snow, though it faces fog. It’s a city that already has many people who don’t own cars, and it was the birthplace of ride-hail, first with the now-defunct Sidecar, then Lyft and Uber. It makes a lot of sense that they both want to be there.
Even so, the city of San Francisco and it’s agencies have not been too happy with the pilot deployments of these companies and recently wrote letters hoping to slow them down. This article examines the conflict between the companies and their city, considering not just the particular points of contention, but also what sort of relationship makes sense here and how to resolve conflicts going forward.
San Francisco doesn’t have the authority to regulate driving. That’s the California DMV. Ride services are under the authority of the California Public Utilities Commission. The federal government regulates the making, importing and selling of vehicles and keeping them safe.
Both Waymo and Cruise started with limited operation permits from the CPUC, but they recently applied to expand those permits. Cruise is operating only at night with members of the public, and a limited part of the city. Waymo operates 24/7 but not in all locations. Cruise wants to move to daytime operations and expand their territory. Waymo wants to go even further, adding highways and tunnels and more.
The city has written letters to the CPUC, asking to slow down approval of these expansions. They are much more bothered by Cruise than Waymo, and not without reason, but they want to press the brakes on both companies.
In the case of both companies they are concerned about:
- Vehicles stalling and impeding traffic (though Cruise much more than Waymo)
- Growing too fast
- Not providing enough data to the city on activities and problems
- Particularly not having enough data on no-safety-driver operations
- Operations in the downtown core at rush hour
- Not having disclosed all the issues that would happen
- Calling 911 on passengers who were unresponsive and just turned out to be asleep, wasting emergency services time
- Doing PuDo (Pick-up/Drop-off) while temporarily blocking a traffic lane
In their protest about Waymo they are a bit kinder, even commending them for a number of their efforts, while still complaining they want to put too many cars on the road at once.
In their protest about Cruise they point out that Cruise has had too many incidents where their cars have frozen on the streets, impeding traffic — sometimes multiple cars. They are particularly upset that a Cruise drove past a fire scene in June of 2022 and drove over a fire hose. A more recent report of another Cruise approaching a busy fire scene describes the fire crew feeling the vehicle was going to interfere with their operations, and so they smashed its window to stop it. Cruise says the vehicle was already stopped and acted properly, and was just waiting for a rescue worker to drive it away —you can read a deeper analysis of all the details of this incident. Suffice to say the city feels that incident represented a serious problem.
Cruise also has had at least one injury accident, and possibly another, though they have not released many details.
The Authority of Cities
Before considering whether or not SF should make these complaints, it should be noted that several of these complaints are understandable and valid. There is no question that these vehicles have made mistakes. The call for more data on the number of mistakes is reasonable. A good resolution on emergency services matters should be arrived upon. The companies have not been very transparent about it either, though part of that will be because they don’t want their mistakes to come back to bite them with regulators, a Catch-22 situation.
At the same time, there is a reason that individual cities are not granted authority over operation on their own streets. It’s not that they shouldn’t care or won’t have valid complaints, it’s that if you want progress, you can’t have thousands of different regulatory bodies requiring data, and paperwork, and appeasement. So the SF departments are going through the right channel when they send these letters to the CPUC, but only to a point.
You would think that San Francisco, which is both the big city north of Silicon Valley and also now a large part of Silicon Valley thanks to the companies that have put headquarters there, would want to be an innovation leader. The reality is, it doesn’t — in fact it’s a little more conservative about these things than average. It is entirely reasonable for a city to take such a stance, and to be, in some views, behind the curve, and in their view, appropriately cautious. However, if a city does that, it’s not going to be a seat of innovation. It will be a follower, rather than a leader.
When Dean Kamen had dreams of the Segway changing how people got around cities, he reached out to many cities to prepare them. San Francisco’s reaction was to be one of the very few places that banned the Segway from its sidewalks, even before it was out. While the city doesn’t regulate cars, it does regulate its sidewalks.
When several Bay Area startups (including Starship, which I was advising at the time) wanted to develop sidewalk-based delivery robots, one city council member in San Francisco took preemptive action before they barely existed to effectively ban them in the city. Both of these bans happened to nascent technology which did not even exist enough to cause a problem, and which did not later cause a problem in other cities. The bans were not to meet a public outcry or solve a problem, but rather just a political expression of caution.
Those who agree with caution might indeed laud these bans. But this is not the pattern of a city that will lead the way. San Francisco can, of course, decide not to be such a city, and it has. The world needs all types of cities and all types of countries. The idea of “jurisdictional competition” where jurisdictions experiment with and innovate on policy in order to allow (or slow) innovation in their cities is part of how the world works.
Indeed, it is not accident, if you look on the Autonomap that most of the action in Autonomy is in the USA and China, with very little in Europe. Yes, these two nations are high-tech powerhouses, but their regulatory approaches are also making a difference. The same is true in trucking.
Waymo already has said it will deploy in Los Angeles this year. While that doesn’t offer the street complexity of downtown SF, and is more sprawled, perhaps it will be a greener pasture — or perhaps not. There are only so many good early candidates. Phoenix actually won the first race with Waymo, which went there before SF or Silicon Valley, after doing most of their early work in the Bay Area.
Traffic is going to get blocked
Many complaints have been seen, both online and from the cities, about incidents where cars blocked traffic. Based on the city’s map, almost all of them have come from Cruise, and only a couple from Waymo, though Waymo operates all day and had a long blockage on a major road during morning rush hour, while all of Cruise’s blockages have been late at night.
In several of these cases the companies could have done better. In particular they had to resolve the problem by dispatching a live human rescue driver to go to the car and manually drive it away. It is definitely not supposed to be that way — both teams have built remote operations centers which allow a human operator to look out through the sensors of the car and give it a human-chosen path out of its problem. They don’t drive the car directly, but give it a plan.
For reasons not always disclosed, the companies have not been able to use this approach. It may be because of failures of the vehicles or the communications link or the remote operator software. In one case, it appears to be a deliberate choice — the company feels that the situation is too complex or risky to allow the vehicle to move itself even with remote human approval, and decides the best solution is to send the rescue driver. This is the conservative decision, in that it eliminates the risk of the car doing something wrong under (supervised) software control, but it can mean leaving a car where it’s not supposed to be for too long a time, and is not a good long term solution except in very rare events.
At the same time, these are pilot projects — small deployments which are there to learn about unexpected problems, and fix them before they scale up to larger deployments. It would not be a pilot if it was expected to go off without a hitch. Those hitches are going to block traffic. What they shouldn’t do is put people at unacceptable risk of being hurt.
Human drivers block traffic every minute on the streets, and nobody notes it on the news. Any drive will encounter double-parked cars, stalled cars and confused drivers. We don’t like it, and it’s mildly unsafe and a violation of the vehicle code in many cases, but we tolerate it. It would be an error to be entirely intolerant of traffic blocking mistakes from robocars in their pilot deployments. We might measure how often humans block traffic and allow the robots at least a similar leniency, but possibly even more.
There is a similar issue around the PuDo question. We all know that cab and Uber/Lyft drivers stop in the middle of the lane for PuDo quite often. If we want to crack down on the robotaxis, we should not be being so forgiving of the human drivers. Particularly late at night.
Part of our ethics of the road include tolerating some problems to help people learn. We regularly find ourselves annoyed behind a driving school car. It has a sign asking us to be patient, and we remember our own youth. We tolerate their issues because this is the only way to turn them into better drivers. We also accept newly minted drivers who just passed their test, out on their own for the first time. These drivers are much riskier than mature drivers, but again, this is how they turn into mature drivers.
With robots, it’s a much more enticing proposition. When a robot makes a mistake, it gets detected and it gets fixed, and soon, all the robots have learned not to have that problem for all time. When a student learns, only that student improves. When a human gets a ticket, other people still keeping doing the violation that earned that ticket. Not so with robots. Indeed, I would wager that when Waymo or Cruise get caught in a mistake, the people at the competing company check to make sure they won’t make the same mistake, and fix it if they do.
This is the boon that robotaxis will provide. Each day they should get safer and better on the road, and very rarely will they regress. Humans backslide all the time. As much as we might like otherwise, there is no other way to truly find all the things that commonly go wrong except being on the road. At first, it was done with a human on board to supervise and that found most of the problems. But some will only show up when the car is empty — such as firefighters getting worried about the empty car and smashing its window.
While Cruise has had more problems, they should be lauded for doing their initial pilot only at night — though they now feel ready to change that, and already have done so for their employees. But operating at night, Cruise has assured that what mistakes they do make are unlikely to snarl traffic. As they expand to the daytime they will be held to a higher bar.
Cities, as one might expect, will be concerned about what they see happen in their city. They will want to do special rules about their experience. This is one reason that cities don’t have authority over this. The decision to be tolerant of these “students” is a public policy decision for a higher level. If the city needs to regulate each individual matter, it is probably not the right city to be the pioneer. It can be Ludberg if it so chooses, but not if it wants to lead the world.
This doesn’t mean that the public, and even officials, can’t point out problems and ask for improvement. Indeed, that’s the fastest channel. The pace of regulation is very slow. The CPUC has said they may not evaluate these applications until May. That’s not the pace that software teams work at. According to a report in the New York Times, Cruise had consultants evaluate their problems blocking traffic back in the summer, and did indeed come up wanting. Cruise states that they have now fixed that problem — before the city even wrote its letter. While companies can’t be trusted to not be selfish, and they will act primarily for their own interests, not the public interest, there are ways to align those two interests — and get problems fixed at the speed of software rather than the speed of regulation. Only if that breaks down do you want to invoke the slower process. Cruise definitely doesn’t want to show up on the media regularly as blocking traffic, or to be dispatching rescue crews.
The issue of 911 calls on sleeping passengers raises a bunch of issues worthy of a different article. Companies have done this because if they delay in calling 911 on a passenger who actually is having a medical emergency, our system will leave them in a world of hurt. When to call 911 needs to be worked out among the various parties, including the courts who will assign liability if somebody has a heart attack and the robotaxi company spends 20 minutes trying to wake them up because they are afraid of calling 911.
This is a new problem that only shows up with a full robotaxi, and it has to be solved, but it’s not the city and CPUC which will solve it. In addition, the solution some might suggest — full time video monitoring of all passengers — risks an Orwellian surveillance regime as we move our travel to robotaxis, and isn’t what we want either. Perhaps passengers can indicate if they want surveillance, or whether 911 should be called or not called (at their expense or risk) if they won’t wake up when spoken to. Or the cars could drive to a nearby paramedic/doctor/facility. (I have long ago proposed that a network of such people who do let their location be tracked via their phones, would be a very handy thing for people having emergencies. No matter where you feel an emergency in a busy city, there’s probably a professional within a 60 second drive, and a defib machine 30 seconds further.)
Our old regulatory regime needs a rethink in the world of robots. Chances are there will never be more than a score or two different companies making self-driving systems on the roads of the state. You can get all the different teams in a room, and just work out good solutions to problems. Only when there is an impasse will it need to escalate to slower regulatory bodies or the courts. You don’t need a 792 page vehicle code — yes, that’s the real size — to deal with people who can’t be trusted or gotten together in a room. Instead you need to define the goals and public interests, and then work to make them happen. Companies can be trusted to primarily look after their own interests, but a fast regulatory regime is very much in their interests.