Land Rover builds cars with two principles in mind: off-road capability and in-car comfiness. When you buy a Range Rover or Discovery, you’re paying for a vehicle that can clamber over boulder-strewn trails and give you a back massage at the same time. So it shouldn’t be surprising that last week the automaker announced it is developing the ultimate combination of these qualities: self-driving cars that can go off-road.
The $5 million project, called Cortex, will give customers “autonomous cars capable of all-terrain, off-road driving in any weather condition.” Now, these won’t be Robo Rovers that can plow through streams and scramble over hulking tree roots—at least not anytime soon. Rather, it’s an early foray into what AVs will look like on off-road terrains.
“We look at customers going off-road and how they use their cars,” says Nigel Clarke, the manager. “Our customers will ultimately want more than on-road autonomy.” For plenty of drivers—like, say, the the Queen of England— that means tackling dirt roads and trails. It’s a generous definition off-roading, but that doesn’t change the idea that those valued customers might want vehicles as capable on the grounds of Balmoral Castle as they are in London.
The problem with taking self-driving off-road is that you give up the predictability of streets designed for cars. No lane lines, no curbs, no reliably stark difference between the road and whatever’s next to it. “Even working out where you can drive becomes more difficult,” Clarke says. To crack it, his team of about 20 people is working with the University of Birmingham (not the one in Alabama) and Myrtle AI, an artificial intelligence outfit based in Cambridge (not the one in Massachusetts).
Right now, the researchers are focusing on radar. For active cruise control in production cars, they use only about 10 percent of the data it returns, focusing only on moving objects likely to be other vehicles. (That’s the same thinking that keeps Tesla’s Autopilot system from seeing stopped firetrucks.) Clarke and his fellows want to use higher resolution radars now coming onto the market and take advantage of the other 90 percent of what they see, so their cars can detect more than fellow drivers on the freeway, with better resolution. That means, however, being able to process gigabytes of data every second.
“The thrust of Cortex is how we can develop algorithms and smart processing techniques to retain as much of that information as possible, whilst making the big data issue containable,” Clarke says. And that means exploring the sort of deep learning that’s at the core of advanced self-driving systems around the world.
The project is slated to last 30 months, so the team’s ambitions are limited. Clarke doesn’t expect to cover millions of miles or master the intricacies of a fully autonomous system in that time. But for an automaker taking an iterative approach to self-driving—gradually implementing more capable systems, model year by model year—getting familiar with advanced tech like deep learning is an important step forward. And for Land Rover, moving forward usually means leaving the paved path behind.
More Great WIRED Stories
- How Facebook groups became a bizarre bazaar for elephant tusks
- Larry Page's flying car project suddenly seems rather real
- PHOTO ESSAY: The trailblazing women who fight California's fires
- Climate change has made zombie ants even more cunning
- The state of federal government cybersecurity is bleaker than you think
- Hungry for even more deep-dives on your next favorite topic? Sign up for the Backchannel newsletter
