A Cruise Robotaxi, with nobody aboard, has rear ended a San Francisco Muni bus on March 23. Nobody was injured and damage to the bus was modest, but more severe to the front bumper of the Cruise. There were no passengers on board the vehicle — Cruise does not carry public passengers during the day, though GM employees are permitted to ride.
Cruise issued a very minimal statement saying:
“Yesterday one of our vehicles made contact with the rear of a Muni bus. No injuries were reported, and there were no passengers present in the Cruise AV. We are investigating the incident and will take follow up actions if appropriate.”
Cruise confirms the vehicle had no safety driver and was in autonomous mode. And while again, no statement is made, it seems very likely Cruise would be quick to tell us if the accident were not the fault of the Cruise, ie. that the bus somehow backed up, or cut quickly in front of the robotaxi. As such it’s a reasonable assumption that the Cruise vehicle rear-ended the bus at a modest speed. Airbags are not visible as deployed (though airbags usually do not deploy if no passenger is in the seat.) Still “made contact” seems an understatement.
It gets worse for Cruise. During the major high-wind storm earlier in the week, a Cruise vehicle drove through “caution” tape closing off a road where overhead trolley wires had fallen. The car then hit the wires, for which power had been shut off.
Both incidents look very bad for Cruise, and they could easily have been much worse if there had been passengers in the vehicles, or if the power lines had been live.
The back of a bus is almost impossible to miss for the sensors in the Cruise vehicle. It could not be more obvious in the LIDAR. If it was stopped, it is possible that a more primitive radar would miss it, but if it had any motion it would be an extremely clear radar target. Computer vision should also do very well on such a target, it being bright and sunny. Even if stopped, for this size target even radar should have done the job. Something had to go very wrong for this vehicle to hit the bus.
Even a complete failure of the main drive computer should not cause such an impact. That computer will have backups, and the backup computer would immediately seek to stop the car in a safe way. If the backup computer had any sensor data, it would have known about the bus, and triggered a hard stop — especially with no passengers aboard.
Absent some special explanation from Cruise, which they are not offering at the current time, this crash indicates a serious failing on Cruise’s part. The language that they will “take follow actions if appropriate” is odd — it seems as though there is no way that a fix is not appropriate here, absent the bus having backed into the car.
The wires and storm
The wires are also a serious failure, though more subtle. If caution tape were to twist horizontal, so it is barely visible to the eye, it would also escape LIDAR and radar. A trolley cable is thicker and should be less likely to be invisible, but it’s not out of the question. If the caution tape was vertical, it should be visible to the vision system, being bright yellow, and also to the LIDAR. Radar is unlikely to see it. It would possibly catch the wire though as a stationary object, it is hard for such radar returns to be reliably processed except with a good imaging radar.
Cruise should, however, have tested this at their test tracks, and in simulator. It was one of the worst storms in recent times for San Francisco, but that bad storms will come is no surprise. All companies should be testing their cars in simulated storms, and also physical simulations at their test tracks. Detecting things like caution tape is also something that should have been tested.
Frankly, it can be argued that with once-a-decade storms, it would be time to have safety drivers. Waymo also had problems, and had vehicles running with no safety driver. Videos show the Waymos running over large fallen branches with significant bumps, though it seems no harm or damage was done. Even so, passengers were surprised and while the vehicles did see the branches and slow for them, it is likely they should have slowed more.
But Waymo did nothing like what the Cruise did. I am surprised these two incidents have not created a “stop the fleet” event, where they temporarily shut down or revert to safety drivers until the issue is resolved. While accidents will happen and you can’t stop the fleet for every one, both these accidents are in the “should never have happened” class. (Yes, one can argue that any accident is in that class, but in reality there are degrees of severity and improbability.) I hope we will see a statement from Cruise about why these didn’t call for such a hold.
The downed-line event could have been prevented with a system, such as I proposed earlier, of communication between emergency crews and robocar companies (and human navigation companies as well) about the location of emergency sites where cars should not go. If there were a means for the dispatch of crews to a downed power line to trigger that area being temporarily blocked on the map, there could be fewer incidents of this type, though testing on the test track is also required.
Cruise needs to improve transparency. They recently reported a million miles of service with no safety driver, claiming “no life threatening injuries” over that period. That’s good to hear but these two incidents also have no injuries but are not a good harbinger. We want to see the full log of incidents if we are to judge how well Cruise is staying safe. Having incidents is not necessarily a sign of failure if they are rare enough, but the anecdotal reports are starting to raise doubt about their rarity.