An SFPD officer drove over a self-driving car. What the viral event means for SF’s future

The autonomous Chevy Bolt from San Francisco’s Cruise was stopped for not having the headlights on. It briefly stopped for police, then took off, crossed an intersection, and stopped in front of a Chinese restaurant, which Cruise later explained as it went “to the nearest safe stop for the intended traffic stop.”

The officer and two colleagues crowded around the vehicle, tried to open the doors, looked in through the windows and shone with flashlights inside. Eventually, according to Cruise and SFPD, the officers called Cruise, who took the remote control of the car. No quotation was issued.

Experts said the incident showed that companies with autonomous cars still have a way to go to find out about human-robot interactions – although some shortcomings could have been remedied with basic common sense.

Situations like this are bound to be more common now that both Cruise, a spin-off from General Motors, and Waymo, the self-driving unit of Google’s parent company Alphabet, drive autonomous cars on California public roads with no one behind the wheel.

Both companies are testing robo-taxi rides for employees and select members of the public in San Francisco, whose winding streets, steep hills and abundance of cyclists and pedestrians make up a famous “bony” traffic training ground.

Robot car companies “try to navigate regimes full of uncertainty and imperfection in the small dances between law enforcement and a vehicle,” said Bryant Walker Smith, who teaches law at the University of South Carolina and is a visiting researcher at Stanford’s Center for Internet and Society. “As futuristic as we think we are, we need to engage with the structures of today and even the anachronisms of the past.”

Cruise has a detailed manual and video guide for first aiders who say they should call 1-888-662-7103 when problems arise. Waymo has similar documentation and a number to call. Both said they are conducting training for first aiders.

But for officers who have not read Cruise’s 24-page document or seen its 19-minute video, its cars obviously lack a clear enough way – perhaps a printed map on the dashboard? – to instruct people outside of them how to reach a human being.

It should have been more obvious how to communicate with the car and / or its human back-up team, said Wendy Ju, professor of information science at the Jacobs Technion-Cornell Institute at Cornell Tech in New York and expert in interaction between robots and vehicles . . She has done consulting work for Cruise.

“The police should not have had to search around to find out,” she said.

Waymo, which has reaped many years of operation in Phoenix, where it has a robust robo-taxi fleet without a driver, said it has developed its interaction with first aiders over time.

A team of remote monitors “will be notified if a vehicle is stopped by police and will immediately roll down windows and communicate with the officer through the sound system of the car,” Waymo spokesman Nicholas Smith said in an email.

Cruise provided a more detailed overview of the incident in response to questions from The Chronicle.

All its cars can identify emergency vehicles by their lights and sirens, it says. In this case, its car spotted the light of the police car and stopped in its lane. Remote cruise operators then asked it to pull over “at what point the AV identified the nearest safe pullover site across the intersection,” the company said.

An electronic tablet in the car’s front console showed a message saying, “First responders should contact Cruise at 888-662-7103,” Cruise said. It did not explain why the officers at first seemed to have trouble seeing that message.

After SFPD officers called Cruise, its remote operator rolled the car’s windows down for officers and “solved the headlight problem,” it said. Cruise said human error caused the headlight to malfunction.

Cruise cars are equipped with two orange buttons with phone icons that passengers can press to talk to a human. But the company says law enforcement should call its 1-888 number instead of using the buttons.

David Sirkin, executive director of interaction design at the Center for Design Research at Stanford’s Department of Mechanical Engineering, said the cruise incident provides a teaching moment – and illustrates the type of interaction that is likely to occur more frequently and should be anticipated by robot car companies.

“I wonder to what extent (Cruise) has considered scenarios where the vehicle has to communicate with people outside the vehicle,” he said.

As cars become more automated, they may need external interfaces with other road users, other vehicles, pedestrians or cyclists, ”he said. “If a car is stopped, or if someone starts yelling at it or throwing a stone at it – how should the car handle it?”

In the Stanford Center’s test of human-robot interactions, it observed times when people bullied the car – shouted at it, stood in front of it to stop its advancing progress, stepped on the brake if they drove in front of it.

“If someone starts abusing a robot car, you might want it to have some sort of escape mechanism,” Sirkin said. “You do not necessarily want a car shouting back at people shouting at it; autonomous cars need ways to assess and possibly de-escalate such situations.”

Smith had similar thoughts.

“One thing that becomes clear is the need to have a less ad hoc approach to all of these interactions: whether it’s with police officers, first aiders, tractor drivers, other road users,” Smith said. “It can not just be that there is a whole pile of 1-800 numbers that police officers are expected to call. There must be a better way to identify who they are dealing with and get in touch with the company. ”

Overall, Smith said: “This is one of the relatively good news. No one was hurt. Cruise did not even get a ticket – how lucky was it? By and large, things worked out. But I say ‘kind of.’ ”

Carolyn Said is a San Francisco Chronicle contributor. Email: csaid@sfchronicle.com Twitter: @csaid