On the road with the Google self-driving car: a recent test drive showed mundane results by human standards, a milestone in automation otherwise. Google plans to automate driving to eliminate (or at least minimize) the consequences of human errors.

Google's ambitious prototypes have docked past 700,000 hours since development began in 2009. The company intends to have working units to be publicly accessible between 2017 and 2020.

The project intends to address car accident statistics related to human error. An estimated 33,000 people in the US die annually from car crashes, and the death toll is largest among people under the age of 45.

Project Director Chris Urmson imagines a world of traffic safety, with accurate, foolproof vehicles hugging the roads.

The technology that separates the Google self-driving car from other sensor-driven vehicles is a topographical map that aids in anticipating objects down the road. It's a layout of the possibilities the car can consider in navigating the urban terrain.

The features incorporated into the topographical map include the height of traffic signs and their placement, the proximity of crosswalks, bends, and the depth of curbs. Lane widths and markings are also factored into the layout.

Automation is complex but still crude by real-world standards, but lead mapping engineer Andrew Chatham states the vehicle is eventually weaned from depending on the topographical, mountain-view map. "We're certainly relying less on the perfect accuracy of the map as time goes on. We're also improving our ability to build the maps."

Information is fed from an assortment of sensors, as well as from a camera mounted on the roof of the vehicle. Data is processed through Google's algorithms, and two numbers are gleaned; the amount of throttle and the turn angle of the steering wheel.

The Google self-driving car can do perfect turns in real-world situations, with variables like incoming traffic and the uniqueness of the road's layout considered (cnet.com).