Why Google’s self-driving car won’t work…or shouldn’t

googlecar1Google’s driverless—or to use proper terminology, autonomous—car won’t work. At least not the way it should. And it’s not a matter of technology as such. It’s a matter of how available technology is used.

Google has a classic case of hubris, and it’s easy enough to understand. After all, it searches thousands—no, millions—of websites to deliver the most appropriate answer to your query (although I suspect that CarBuzzard doesn’t come up at the top of searches nearly often enough) via massive banks of data stored hither and yon around the world on massive computers.

And that’s fine for static web searches. The computing is centralized and top down, sending out bots to crawls websites and report back to headquarter. And that works.

But the fluid nature of cars traffic is different. As Andrew Chatham, mapping lead for Google’s self-driving cars project told Automotive News that the cars’ systems work by asking four questions: Where am I? What’s around me? What will they do? What will I do?

The answer to the first question is GPS and 3D maps, which tells the car what the world looks like when it’s empty of moving objects. The second analyzes the differences between the data stored—wherever it is stored—and what’s detected by the various cameras. “It’s the job of the software to figure out how the world is different from that expectation,” said Chatham.

From there it’s a matter of software to tell the car what to do: accelerate or slow down, turn right of turn left.

It’s a complex task, but systems that do that are already in the field. Automatic braking is commonplace as options on premium models, and Volvo’s system can recognize cyclists, pedestrians and predict what they’ll go. Cars that “talk to each other” is the next step, although this requires a common language—Microsoft versus Apple, anyone—as well as developing appropriate and coordinated responses. Who, for example, yields to whom?

But where there are technical systems choices—VHS versus Beta—can lead to a less than optimal victor, can be outdated by an emerging technology. The DVD player has replace tape. And does anyone remember ZIP drives? The autonomous automobile is in its infancy, but as someone wise once noted, some of our most important decisions are made when we are teenagers.

Even if Google can provide data in a split second, that may not be fast enough. Cars won’t carry the world of highway data on board, the data will have to come from somewhere, and that somewhere has to be satellites. And as anyone who has satellite radio knows, all it takes to lose contact is a hill or tall building, and it has a limited amount of bandwidth for data—and therefor music quality—transmission.

Of course, Moore’s Law says that computing power doubles every eighteen months, and that could be the data problem solution. However theoretical physicist Michio Kaku, as quoted in Time Magazine, sees that trend leveling out completely by 2022. But in 2012, Google co-founder Sergey Brin set a goal of having the technology commercialized within five years. And that’s only 2017.

Getting hardware to be where it will need to be will be hard enough, but with only 2,000 miles of roads mapped sufficiently for its current autonomous automobile to work, Google has a long way to compete the four million miles of public roads in the U.S.

So is detailed centralized data the right way to go? Or would self-driving cars with more extensive “local” processing, reacting not so much to details from Google central but providing the basic elements of where the vehicle is, and let the car do the rest, much like humans do with GPS. Humans don’t address every data point available, just those that are relevant. Most of the time but not all, we’re pretty good at it.

And we4’re smarter than the average bird brain. The starling however performs something almost miraculous. In swirling masses of birds called a murmuration, the flock twists and turns like liquid smoke, and no bird ever collides with another. Research has shown that each bird only has to track seven other of the other birds. The flocks have no leader, just a form of distributed real-time computations.

In the Fifties, the space program of the United States was “space planes,” including the X-1 through X-15. That changed, however, when the USSR put a small satellite into orbit with a rocket. To keep up with the Russians, our focus changed from planes to rockets, essentially an expensive and inefficient way to catch up quickly in the “space race” using brute force.

We shouldn’t allow brute force, not to mention Google tracking us wherever we travel, to get us to the autonomous car on the road. We shouldn’t do the same thing with the self-driving car.