With Google’s self-driving car project now having racked-up 700,000 miles – most of them on freeways – the company is now focusing on teaching the cars to cope with the far greater number of variables found on city streets …
To me, the most impressive elements in the video are the railroad crossing and cyclists. For the railroad crossing, the car knows to wait until the road on the far side of the tracks is clear before it begins crossing. With the cyclists, the car detects the arm signal and waits for the cyclist to move out, also detecting cyclists approaching from behind at a junction.
The car uses a spinning bucket on the roof containing 64 laser emitters to construct a 3D model of its surroundings, supported by a separate radar system. The main screen in the video shows how the car’s software interprets the complex mass of data needed to safely navigate city streets.
Atlantic Cities journalist Eric Jaffe was given a ride in the car on city streets, and reports that project head Chris Urmson believes that although the problems are significant, the goal is achievable.
To grossly simplify [freeway driving], you follow the curve and don’t hit the guy in front of you [...]
The complexity of [city driving] is substantially harder. But basically over the last year we’ve come to the conclusion it’s doable, and that this intuition we had about making a vehicle that was fully self-driving was correct. That it was possible. That we actually think we can make one that really is safer than human driving.
The car is designed to operate in a fail-safe mode: if it can’t be confident of safely proceeding, it waits. This happened during Jaffe’s demonstration drive, when the car detected cones at road works but couldn’t work out how to navigate past them, so sat and waited for the human driver to take over. The car is fitted with a big red emergency stop button, but in 700,000 miles of driving, no-one has ever had to use it.
The full article on Atlantic Cities is a really interesting read.