To end distracted driving, MIT figures out how people really drive

Lawmakers and parents like to talk about “driver distraction,” but it's not a simple idea. There’s no on or off switch for driver focus. Attention, like so many things, is a spectrum, and it combines many elements.

“Most of the research in the past has been either visual, audible, or haptic—they haven’t been combined all into one," says Douglas Patton, Denso's head of engineering.

In 2012, government-sponsored researchers rigged up 2,600 regular drivers’ vehicles with cameras and sensors in six states, then left them alone for more than a year. The result is a large, objective, and detailed database of actual driving behavior, the kind of info that’s very useful if you want to figure out exactly what causes crashes.

The MIT researchers and their colleagues took that database and added another twist. While many scientists looking to crack why a crash happened might look at the five or six seconds before the event, these researchers backed it all the way up, to around 20 seconds beforehand.

“Upstream, further prior to an event, we begin to see failures in attention allocation that are indicative of less awareness in the operating environment in the crash events,” says Bryan Reimer, an engineer who studies driver behaviour at MIT. In other words: The problems that cause crashes start well before the crunch.

It all comes down to eye glances. Sure, the more time you spend looking off the road, the likelier your chance of crashing. But the time you spend looking on the road matters, too. If your glances at, say, the texts in your lap are longer than the darting ones you make back to the highway in front of you, you gradually lose awareness of where you are in space.

Usually, drivers are pretty good at managing that attentional and situational awareness, judging when it's appropriate to look down at the radio, for example. But smartphones and in-car infotainment systems present a new issue: The driver isn’t really deciding when to engage with the product.

The algorithm that researchers tested in this paper—one called AttenD, which dates back to 2009—turns out to be pretty good at predicting when crashes happen based on what drivers were doing in 20 or so prior seconds. That means that maybe, one day soon, scientists could use this kind of math to build and then test products that are safe to use in the car.

Read the full story here.