Last week’s “exclusive” in the Guardian claiming to “confirm Apple is building self-driving car” raised quite a buzz. Much of that buzz was skeptical, with many pointing out that the facts failed to support the Guardian’s conclusion.
The logical leap that Guardian made was that an Apple engineer’s interest in the GoMentum Station vehicle test track confirmed Apple’s driverless car program. This is too big a leap, as a range of Apple car-related aspirations—self-driving or not—might have use for such a test track.
Chris Urmson, Director of Google's Self-Driving Car Project (Photo: Bret Hartman/TED)
Let’s assume, however, that the Guardian is right and Apple does have a driverless car ready for testing. (This is possible, as Apple has hired many automotive engineers, including the former CEO of Mercedes Benz’s Silicon Valley research center.) What would that say about the relative state of Apple’s driverless car?
It would tell us that Apple is millions of miles behind Google, and falling further behind every day.
As one of the few companies in the world richer than Google, Apple can match the cars, sensors, processors, navigational systems and other pieces of hardware that Google might deploy. It can replicate the sophisticated maps that Google has compiled. It will have a very hard time, however, catching up with Google’s on-the-road learning.
Actual road miles are critical because driverless cars learn to drive like humans do—through experience. Sophisticated hardware tells the car where it is and what the other things around it are doing. The actual driving, including identifying those other things, predicting what they might do, and handling all possible situations to safely get the car to where it needs to go, depends on very sophisticated machine learning algorithms. Those algorithms, in turn, depend on analyzing real-world driving situations.
Google certainly has a ton of such data. Chris Urmson, director of Google’s Self-Driving Car project, has publically stated that its fleet of 20+ self-driving vehicles has driven more than 1.7 million miles, with nearly one million of those miles in self-driving mode.
What’s more, Google is accelerating its learning. It recently expanded its testing to Austin, Texas. It is about to launch dozens of built-from-scratch prototypes in the Bay area. All told, Google is averaging 10,000 new self-driven miles a week, mostly on real-world city streets—not on controlled test tracks. And, it is using its big database of driving cases to run three million miles of simulated driving a day.
Compare Google’s millions of learning miles with the Guardian’s optimistic scenario—namely that Apple is “scouting for secure locations” to test its driverless car. In other words, Apple is just getting ready to get started.
Even as Apple readies its driverless car for GoMentum Station’s well controlled, 5,000-acre 20-mile test track, Google is fielding dozens of self-driving cars onto real roads.
Google’s real-world advantage means that its cars are learning to deal real-life driving scenarios involving its passengers, other cars, taxis, trucks, cyclist, pedestrians, emergency vehicles, road construction, road makers, traffic signals and innumerable other factors—including pushy drivers and even people in wheelchairs chasing ducks.
And, unlike human drivers who must rely on their own experience for learning, Google’s cars will learn from every Google cars’ experience. That means that the more cars Google puts on the road compared to its competitors, the greater its learning advantage.
As I have previously written, millions of lives and trillions of dollars depend on the successful development of driverless cars. That is great incentive for Apple and other companies to challenge Google’s lead.
Google grabbed that lead by thinking bigger, starting small, and learning faster. It is not an insurmountable lead; but, it won’t be overtaken from secret test tracks.
Unlike consumer electronics, like Apple’s iPhone or Watch, driverless cars cannot be created in stealth mode and then unveiled in full form.
The time to start taking Apple, Mercedes, Nissan, Tesla, General Motors or any other potential driverless car maker seriously is when they have the confidence to leave their hidden test tracks and openly log the millions of road miles necessary to teach their cars how to drive.
Also on Forbes:
