Tesla’s robot car handles Zurich smoothly, but what happens when it crashes?

From the outside, the white Tesla looked like any other car in Zurich’s evening traffic. Inside, however, the steering wheel turned by itself, the accelerator moved without a driver’s foot, and the vehicle braked smoothly when two pedestrians approached a crossing.

The women appeared on the car’s display as pale silhouettes among the bicycles, cars and trams around them. They looked toward the vehicle and smiled in thanks, unaware that software had made it stop.

A Tesla employee sat in the driver’s seat with her hands and right foot ready to intervene. The company offers the trips under the label “full self-driving (supervised),” allowing passengers to travel around Zurich for 45 minutes, ideally without intervention from the employee. The demonstration showed how close the technology is moving toward autonomous driving.

The system uses the Tesla’s eight exterior cameras and processes video, navigation information and large volumes of movement data. Tesla says it can also recognize sirens and hand signals, including gestures telling the vehicle to continue or stop. During the demonstration, the car drove across the city to Zürichberg and back with almost no human input.

There was one intervention. On a bridge over the Limmat, the car would not change lanes because a cyclist blocked its path. The employee took over briefly, pressed an unobtrusive button on the steering wheel, and returned control to the software.

That smooth demonstration leaves a harder question for Switzerland. If a robot car, or a more autonomous version, causes or becomes involved in a serious crash, can authorities reliably determine whether the person or the system failed? Experts warn that the country lacks an independent body to secure data after such accidents and identify software errors.