Where’s my self-driving car?

In 1967, Popular Science magazine declared that cars would be self-driving by, at the latest, 1985.  Their vision was of cars that were driven manually for the final mile at either end of a journey, but which were guided by electric railways for the bulk of the trip:

“You gulp the last of your coffee, wipe the egg off your chin, and dash for the door. In the driveway sits a vehicle about the size and shape of a Volkswagen. Beside the door on the driver’s side is a handleless hatch. Beneath the car, unseen, are four flanged wheels of smaller diameter than the car’s tires.

As you slide away from the curb, the sound of the electric drive motor hardly rises above a whisper. A few blocks from home, you steer the car into a special lane, and pull a lever under the dash. The front wheels lock in straight-ahead position. Simultaneously the side-hatch door slides back and an electric third-rail folds out. It makes contact with a power rail, the flanged wheels roll onto the rails of a track, and your car accelerates at a controlled rate of 0.3g. You twirls a dial until you see “5th Street” appear in a small window. Seconds later, as your car enters a main guideway at exactly 60 m.p.h., you open the paper and scan the news. (via Boing Boing)

This was the 1960s, when you could put a man on the moon within a decade.  You’d think that the Americans could manage a simple extensive network of national, regional and local automated road/railways.

Turns out, this system had not quite been completed by 1985.  It’s not at all clear why this dream failed to come true.  But whatever the reason, the engineers at Google have been left wondering, “where’s my self-driving car?”  So they’ve built some of their own.  These cars are far more fascinating than the rail-cars of the 1960s.  These cars drive themselves on normal streets and motorways.  In amongst lots of manually-driven cars.  And pedestrians.  And San Francisco’s stoned cycle couriers.

All the news coverage of the Google Car fawns over its radar and cameras and obvious physical equipment, but far more interesting must be its software.  You can’t program a car with simple rules to accommodate unpredictable human hazards any more than you can create a driver by giving an idiot a car manual and a copy of the highway code.  How do you program a car to know that it needs to be more careful around pedestrians who are young and who are running and when the pavement is next to the carriageway and not segregated with railings and when there is another pavement opposite and there’s a playground nearby and there are parked cars and…?  There are a mere 306 rules in the highway code, but there are an immeasurable number of scenarios that one might encounter on a road.  So the Google Car must surely have used some very clever “machine learning”, and the 14,000 miles on the clock are presumably human-driven, with the car watching and sensing and learning how to drive — how a driver adjusts to other cars and other people, to the weather and the sound and feel of the road, in millions of situations and combinations.

It must be some of the most advanced computer science ever, with some of the most brilliant minds in the world working on it.  Google say that it could be available to consumers in eight years.  And there’s no reason why it shouldn’t be.  It’s a brilliant piece of work.  Just like the self-driving rail-car before it.