July 16, 2015
“We’re pretty convinced this technology is going to come to market,” he said.
Google’s driverless cars are now doing 300 million miles of testing in simulators every day. And in the real world, they are reacting beautifully to unexpected situations: a woman in a wheelchair chasing a duck in circles in the middle of the street; a bike running a red light; and a little boy in a toy car.
The technology works by mapping its surroundings, in increasingly complex ways. These days, the cars can not only interpret the shape of roads, other cars, and pedestrians, but also cyclists and traffic cones. They can distinguish a police car and a school bus; they can interpret a cyclist’s hand signal when merging or a policeman’s gesture to stop.
The driverless car software analyzes the probable path of each object, predicting where they will go and how fast. As the program advances, Google is accumulating more and more data on new situations and new objects (like those ducks). Cars can also share data in real-time on construction zones, detours, and more.
In his TED talk, Urmson argues that we need a driverless car approach, not the “driver assistance” approach taken by companies like Tesla. The idea behind driver assistance is to keep giving more and more help to drivers – like emergency braking and alerts when you drift out of your lane – until cars eventually become driverless. But Urmson believes that incremental improvements won’t get us there soon enough – we need a radically new way of thinking.
He makes the case that we don’t have time to waste – 1.2 million people are killed every year on the road. Those with certain disabilities like blindness can’t even drive, restricting their freedom. And together we waste 162 lifetimes per day in the US on commuting.
Watch the full talk here:
Did you like this article?
Get more delivered to your inbox just like it!