SF Fire Dept. Says Cruise Robotaxi Did Not Yield Properly And Delayed Fire Truck

According to reports, a San Francisco Fire truck on way to a fire was delayed by the combination of a garbage truck blocking the lane and poor performance by a Cruise robotaxi in the oncoming lane. This is the 3rd recent incident between the Cruise robotaxis and the City of San Francisco, which may interfere with their plans to expand their operating permit in the city. The situation is, as usual, complex.

In April, in the Parnassus Heights area, at 4am, a fire crew was on the way to a fire when its lane was blocked by a garbage truck. A Cruise vehicle with nobody in it was moving in the oncoming lane. According to Cruise, their vehicle detected the fire truck and as it is programmed to do, pulled to the right and stopped (avoiding blocking any intersection) and summoned remote assistance. However the oncoming lane was not wide enough for the fire engine to pass, so the garbage truck driver got into that vehicle and got it out of the way.

The fire truck was delayed only 25 seconds, which is probably not an unusual thing to happen, and so the SFFD is probably overstating the seriousness of this incident, though it points out the fire caused property damage and minor injuries and it’s understandable that every second counts. The garbage truck was the main cause of the blockage, but this is what those trucks do. A more interesting question is how could they have done better and what would have happened if the garbage truck had not moved?

The Cruise vehicle found itself in a predicament. Parked cars prevented it from pulling off to the right (as it is programmed to do) sufficiently to clear the lane for the fire engine. This was right at an intersection, so to back up to clear the path would have involved backing up into an intersection, something the vehicles is generally programmed not to do. While Cruise vehicles can reportedly back up, this is not a situation where they want to. Cruise declined to say whether it could have eventually backed up into an intersection (on its own or under command from remote assistance) or not. SFFD seems to feel a human driven car could have pulled this off, though in the end it was not necessary since the garbage truck got out of the way.

Another option might have been for the fire truck to back up and somehow signal to the robocar that it should move forward to clear the lane. Such a signal would have to be by hand signals — which the remote ops center people could recognize — or through an interface given to the fire department. Cruise maintains a phone number that emergency workers can call, and it displayed on the vehicle in situations like these. If the fire crew (or their dispatch) had the number, they could call it quickly.

This is not an “edge case”

Some might say this is a rare “edge case” that you have to expect will be trouble. For this, that’s not the case. Encounters with emergency vehicles are high on the priority list for all serious teams. They have not just thought about them, they have thought deeply about them, writing handbooks and training both their own staff and emergency crews.

In particular, all major teams test their vehicles extensively in simulator, and they should have very large numbers of varied scenarios in which to test this sort of situation. Cruise declined to comment on this, but I would expect them to have tested situations like this in simulator, to look for and fix any situation where it might block a fire truck. In simulator you can test this situation happening literally in every space of every foot of a city, and find and think about the ones that don’t work.

What’s true is that most robocar development doesn’t spend nearly as much effort on backing up. Backing up into an intersection isn’t something anybody is comfortable with and is more of an edge case. Many designs have the best sensors facing forward, and have more limited sensors to the rear — but enough to do low speed operations. Backing up into cross traffic is something many humans can’t do, which is why some cars have rear cross traffic alert sensors, mainly for use in parking lots and exiting driveways.

If one has to speculate, Cruise may not have spent enough time on the idea of backing up into an intersection. Cruise’s project is currently unfinished. The fact that it still does pick-up an drop-off without pulling out of the lane, with the City of SF has complained about, is a big one. It is normal and expected for a robocar project to not be complete while it is testing, and to leave some problems for later if they won’t cause too much trouble. Cruise might have underestimated how much trouble they will cause. Critics may also be overestimating the trouble as well. Uber drivers drop people off in the middle of the lane all the time. And some (but not all) human drivers might be scared to back up into an intersection too.

Staying legal

Robocar developers are strongly motivated to stick to the letter of the law. The letter of the law on emergency vehicles is to pull over and stop, which is what the Cruise tried to do, though it could not because of parked cars. It’s not clear that backing up into an intersection is within the letter of the law, but it is within the spirit of what you should do to let a fire truck get past. The law also, however, forbids the garbage truck or anybody else from double parking and blocking the lane in a way that can’t be cleared quickly. (ie. the driver is supposed to stick around to move the truck in just this situation.) As such this situation is supposed to have a resolution already in the law, if not a perfect one.

This is Cruise’s 3rd reported problem. Earlier, a Cruise car was pulled over by police for not having its lights on. In addition, there have been complaints about how the cars will stop in the lane for pick-up and drop-off. Cruise is applying to be able to charge for robotaxi rides. It will also soon want to expand its service outside of the 10pm to 6am night window they currently operate in, which they chose due to the simpler nature of driving on the quiet streets. The public utilities commission may be bothered by the various incidents reported above, including the one with the fire truck.

I suggest they should not be too bothered. None of the problems have been so serious that you would want to slow down development of live-saving technology like a robotaxi because of them. Robotaxis have to start clumsy and get better, that’s the only way to develop them, and real world experience on the road is the only way to find and fix these problems. Nobody has been hurt — as noted there are many things we accept that might delay a fire truck in the world — but tons of people are being hurt by the human drivers that will some day be replaced by these robotaxis. The sooner the robotaxis deploy, the sooner those people stop getting hurt and killed. A few odd run-ins with police and fire is nothing compared to lots of future deaths and injuries, even though we won’t know who those specific people are. If the vehicle had resulted in a major delay, and in a way that no human driver (not just the best human driver) would have caused, that would be another story.