Here's what two million miles of autonomous driving has taught Google
Google's self-driving vehicles have now driven over two million mi ( 3.2 million km) on public roads and, as you might expect, the first million looked very different than the second. Freeways, light traffic and simple intersections have given way to complex city streets, advanced road craft and social cues.
Google first revealed that it was working on self-driving tech back in 2010. At the time, its autonomous vehicles had covered just 140,000 mi (225,300 km), but had already been able to negotiate San Francisco's tricky Lombard Street, cross the Golden Gate Bridge and circumnavigate Lake Tahoe.
In an article yesterday, Google's head of self-driving tech Dmitri Dolgov explained that these sort of achievements came pretty quickly. "That's because it's relatively easy to master the first 90 percent of driving where you're traveling on freeways, navigating light city-street traffic, or making your way through simple intersections," said Dolgov. "But to create a truly self-driving car that can do all the driving, we knew we'd need experience in more challenging and interesting situations."
The more the cars have driven themselves around and uncovered a range of situations, the more Google has learned. Dolgov suggests that the second million miles has helped Google's cars learn how to deal with even out-of-the-ordinary situations, with a unicycling couple and cars driving the wrong way among those it has encountered.
To gain the right kind of experience - that so-called final 10 percent - Google has its cars spend most of their time today on what Dolgov calls "complex city streets." In these environments, they are said to have become better at detecting obstacles – such as emergency vehicles, construction zones and closed lanes – and responding accordingly.
The cars are also said to have become better at understanding the nuances of human driving and predicting what other drivers are likely to do, based on what Dolgov characterizes as the "delicate social dance" and "silent conversations" of road users. Here, he is referring to the subtle ways in which we indicate our intentions to each other – such as by positioning ourselves to change lanes – that may be obvious to human drivers but much harder for autonomous vehicles to recognize and interpret.
In turn, one of the skills that Google's cars have learned is how to drive more smoothly. This makes their very movement safer and, of course, improves the experience for vehicle occupants. In addition, it communicates to other road users that the "driver" of the car is competent and in control.
This last point is particularly prescient given the news that California has revised its laws to allow self-driving cars without pedals or a steering wheels to be tested on certain roads. Google had previously been put out by the state's insistence that autonomous vehicles must have an operator able to take control if need be. As the rules are gradually relaxed, though, and the likes of Google continue to clock up autonomous miles, the once far-fetched idea of reading the news while your car ferries you to work becomes ever closer to reality.
Source: NewCo Shift