Waymo And Cruise Push Back In Battle With San Francisco

The annual TRB ARTS self-driving conference is the oldest conference in the field, and it took place in San Francisco this week. The hot topic was surely the brewing battle between the city of San Francisco and the two companies doing robotaxi pilots in that city, Waymo and Cruise. While there was no direct debate between the parties, the conference was opened by Jeffrey Tumlin, head of the San Francisco MUNI transit agency which included a fair bit of complaint about problems with the robotaxis. The reaction from Cruise and Waymo speakers was less confrontational, but nonetheless included some smoke from the battle.

At the same time, Cruise this week ran full page ads in major newspapers describing the bad safety record of human drivers and stating that Cruise vehicles were doing much better. Waymo has said this in the past, and earlier in the week released a study they did of just how much it is that people speed. While everybody knows that speeding is very common, Waymo vehicles are always watching the roads in detail and were able to quantify that typically half of all cars are speeding, some doing as much as double the limit. Waymo and Cruise cars do not speed — though there is considerable debate over whether they should at least try to match the typical speed of traffic, even if that means speeding. At present, companies are wary of programming their vehicles to speed the way people do, even if that means better road citizenship.

San Francisco is frustrated by various incidents where robotaxis had stalled on the streets, sometimes blocking transit vehicles, or had bad interactions with emergency crews, delaying them on their way or getting confused at emergency scenes. Initial reports of these incidents were extremely rare, but the city claims the numbers have increased a lot recently, and they demand that the companies provide data on just how often things are happening.

The real frustration, though is that the city does not have authority to regulate the roads — that belongs to the state, including the DMV and the public utilities commission. General feeling is that allowing each city to set its own rules of the road and regulation of services on the road would result in an unworkably complex regulatory regime. As such, the city has been limited to writing letters to the California PUC, asking them to scale back robotaxi operations, and to deny the requests of Waymo and Cruise to expand their pilot service. (At present, Cruise only runs late at night and Waymo is not allowed to charge fares, among other things.)

Cities are not without power, though, and at a recent council meeting, they denied Waymo a permit to install employee parking in a new facility, with speculation this was out of spite rather than because of a problem with the parking lot. Waymo and Cruise can’t remain at war with their host city.

Waymo staff, including safety research director Trent Victor who spoke later in the event, expressed frustration with the claims being made against them. Victor has published many safety papers outlining the very strong — indeed superhuman — safety record of Waymo vehicles (particularly in the Phoenix area) and feels some of the claims circulating are “deliberate disinformation” though he declined to accuse any particular party of t his. He didn’t have to — the California PUC itself accused the city of poor analysis, listing crashes where the robotaxi had been rear-ended while not moving, and not taking into account the higher crash rates for human drivers on urban streets in their complaint.

In Tumlin’s address there was a striking example of the disconnect which occurs between developers and most other parties, including a transit agency official like Tumlin. Tumlin described a number of incidents which concerned him, and of course these were naturally all things from the past and present. As is to be expected, he takes the problems he sees today and extrapolates them as a feared future. Software developers are indeed concerned (almost exclusively )with what their vehicles will do in the future, in particular any issues which they don’t know how to resolve. All the incidents named by Tumlin will be well known to the teams making the cars, and be bug tickets that are already fixed or which are in the process of being fixed. The pilot testing phase is there to find such problems, after all, and to learn about them and fix them. Tumlin spoke about many of them as though they were inherent defects that won’t be fixed, and thus would be a reason to stop or slow down testing. In fact, the whole point of the testing is to find eliminate those problems.

That does not mean that a vehicle might not get on the roads too early, when it has too many problems to be acceptable. There is worthy debate to have over just what level of problems can’t be tolerated. Tumlin likened the vehicles to drivers with learner’s permits in order to criticize their performance, but in fact that is the purpose of learner’s permits – they are the means by which the world turns unskilled teens into passable drivers. Without them, nobody would ever get on the road.

To be fair, there is one big issue which the teams have been slow to fix, which Tumlin referred to as the vehicles becoming “bricked” on the streets. (The vehicles are not bricked in the sense a computer programmer would use the term, when they are indefinitely as unresponsive as a brick, but they definitely are an issue.) In many cases, the companies are resolving problems by sending a human rescue driver to extricate the vehicle rather than having remote operators give the vehicles guidance on how to move on, as they do when they encounter rare road situations. The companies have been loathe to give details about these incidents, but it appears they are doing this out of an abundance of caution. If they are not 100% sure that the remote advice approach will be safe, they choose to leave the vehicle stalled until the rescue driver can resolve things. Human driven cars don’t have such problems. It may be that the teams need to work on being able to remotely resolve almost any problem, possibly even using full remote driving, something that some companies do but which Waymo and Cruise wish to avoid.

It is also apparently the case that Cruise is having a lot more of these problems than Waymo, though part of that is explained by the fact that Cruise was operating more vehicles in SF, and it was until late last year their only territory. Waymo still has most of its miles in the Phoenix area, which is easier to drive.

The city’s demand for more data is a reasonable one, though. It does make sense for the companies and the state officials who regulate them to work together with the cities to figure out a good data reporting regime so that people can get hard data on what is happening, define what is acceptable and what isn’t, and work to stay within those limits. Companies are naturally wary of releasing data which could either burn them later or help their competitors, but it seems likely a middle ground can be found. San Francisco should make a case for what level of issues should not be accepted in a pilot project, and the state should use that to define what sort of data are needed, perhaps to be examined in confidence by a 3rd party. Both Waymo and Cruise have declared that they are driving safely and not causing particularly high disruption to the streets, and should be happy to confirm that. At the same time, we accept a lot of disruption on our streets from human drivers (including those with learner’s permits) and should accept at least that much from robotaxi pilots, and probably even a bit more, given the promise they have of what they will do when they get better and start driving at scale.