This winter is going to be interesting.

Flat landers or rude big city easterners.

Snark...

Seriously though, as a native to the area I found it strange that Uber chose Pittsburgh as the test city. The issues to over come; geography, street size, tunnels, bridges and random one way street placement make it a unique challenge.

It's not easterners humping self-driving cars, it's the CA urban folk, where it hardly ever even rains.

The harder the test city, the closer to ready they will be when they are ready. Google is softballing their development compared to Uber.

Computers are great at some parts of driving, and suck at others. For most driving tasks, computers are better than we can ever hope to be. They can have omnidirectional unblinking awareness that can maintain safe distances with pitch perfect lane placements. That will (almost) never dangerously cut in front of an overtaking vehicle, and can provide the best lane selection for navigation. But, outside of some formulas and algorithms and sensory inputs, they have NO clue, no understanding of what it means to be "in motion".

Meanwhile we can look at the world around us and usually know what things are, and how they affect what we are doing. From the very earliest stages of biological life, our species has evolved to be good at moving. Not only through vision, which is easily tricked or in some cases unavailable, but through our equilibrium. Combined with reasons to relocate and raw survival instincts, even the most developmentally disadvantaged understand what it means to be "in motion".

It is the edge cases that make designing self driving cars such a challenge. As a simple example, the effects of time of day. Changing lights mean changing shadows and changing colors, which makes quick and simple object recognition MUCH more difficult. Heading east at sunrise, or west in sunset? The cameras may be even more blind than you. Rain and snow and the fading of time take away lane lines. Is that a shadow or a pothole? Hurry up, you only have 20 milliseconds to figure it out and decide what to do. Do you hard brake, steer a little, steer a lot? Can you straddle it? Are you in traffic where a hard maneuver will create a disaster?

Can the SDC see cars a half mile or more braking or maneuvering, and react long before it even needs to be an issue? What exactly is a particular pattern of speeds, brake lights, and lane changes telling us? Even poor drivers are better than computers at that.

I believe the BEST application of this emerging technology would be to help alert drivers be more alert, rather than trying to replace them.