Speaking at the Automated Vehicle Symposium Wednesday in Ypsilanti, about 30 miles west of Detroit, Chris Urmson, director of Google’s self-driving cars, showed a variety of scenes the cars have confronted on surface streets around the company’s Mountain View, Calif., headquarters.
One case featured a duck scampering across an intersection followed by a woman in an electric-powered wheelchair going back and forth, in apparent pursuit of the duck.
“It doesn’t matter how long you gave me, I never would have come up with that scenario. There are no rules. The DMV has nothing in its handbook,” Urmson said.
The car must have the ability to determine “this is weird. I’m just going to chill out here and let that all play out.”
But human drivers might watch the path of the woman and the duck and proceed if they think they see a safe opening. The Google car will simply wait until its cameras and sensors don’t notice any impediment.
Urmson showed another example where three lanes of traffic were about to proceed from an intersection as the light turned green. The Google car was in the far right lane. But coming from the left was a bicyclist who was determined to shoot through the intersection despite a red light in his direction.
The first two human-driven cars moved forward forcing the cyclist to swerve toward the middle of the intersection to avoid them. The Google car held its position because one of its sensors picked up the cyclist’s speed and concluded it was about to bolt through the intersection.
But this extra degree of caution, at this stage in the technology’s development, creates another level of risk.
Google has begun reporting accidents involving its vehicles. It turns out they have been involved in 14 minor accidents over more than 1.8 million miles of testing. But the company insists that the self-driving cars did not cause one of the collisions.
“There’s been a lot of noise recently in the press about the fact that our vehicles have been in collisions,” Urmson said. He then showed a video of one the recent accidents where the Google car had to stop because of other vehicles stopped in front of it. Meanwhile the car behind it didn’t notice the stopped vehicles and rammed into the Google car.
That illustrates one of the unsolved challenges before autonomous vehicles enter the mainstream of our mobility networks. Yes they can be programmed to be as safe as possible given the immediate space around it.
As Urmson himself said, “People are going to be people.”
Ideally human-driver vehicles will be equipped with enough sensors, cameras and software to override human mistakes. But we aren’t there yet.
©2015 Detroit Free Press. Distributed by Tribune Content Agency, LLC.
