Friday, 21 July 2017

New Company Crowdsources Mapmaking to Pave the Way for Self-Driving Cars

Leave a Comment
http://ift.tt/2tn30Ko

Lvl5-2

Do you earn a living by hustling for ride-hailing companies or driving delivery gigs? There may be a new way to add a few extra dollars to your bottom line while helping to lay the groundwork for autonomous vehicles.

Mapping ever-changing road environments is one of the trickiest aspects of preparing self-driving vehicles for the road, and there’s a new company that aims to do just that by crowdsourcing location information from the cameras on drivers’ smartphones. In return, drivers can earn 2 to 5 cents per mile.

Founded by a team that includes two former engineers who worked on Tesla Motors’ Autopilot semi-autonomous feature, Silicon Valley startup Lvl5 is recruiting drivers to download an app called Payver. With the app running and smartphones mounted on a dashboard, the company’s engineers can use computer-vision software to distill a high-definition portrait of the vehicle’s surroundings.

“Every self-driving car needs a map,” said Andrew Kouri, Lvl5’s co-founder and CEO. “It’s more than just a Google map—it needs to be a high-definition map of everything on the road, where sidewalks are, crossing lights, and everything else you can imagine that would help a car know how and where to drive. What it comes down to is that nobody has these maps at scale right now.”

“What it comes down to is that nobody has these
maps at scale right now.”
– Andrew Kouri, Lvl5

That’s not for lack of trying. At least a half-dozen companies, including Mobileye, Civil Maps, Here, and TomTom, are vying to provide these maps for autonomous vehicles. In a move that caught the attention of the industry and demonstrated the value of these maps, Intel acquired Mobileye in a deal worth $15 billion earlier this year.

Kouri helped integrate Mobileye’s cameras into the Autopilot system during his days at Tesla, and in building his own company, he is following a similar approach that favors camera-based maps and eschews those made with lidar sensors. While a number of autonomous-vehicle developers rely heavily on lidar units, he said they cost too much and remain unproven in real-world environments.

Kouri claimed the company’s camera-based software, which snaps a frame every three feet, can provide the same accuracy as lidar.

This image is an intersection in San Francisco generated from information taken by the Pavyver app after two passes through the intersection.

This image is an intersection in San Francisco generated from information taken by the Payver app after two passes through the intersection.

“We would totally buy lidar if it made sense, because it works so well for autonomous cars, and it’s a great sensor,” Kouri said. “The problem is, it hasn’t been tested in Michigan winters or gone through carwash cycles. These are questions if you are mass-producing an autonomous car—you’d better know that a $5000 component is going to work. So there’s that, and nobody is producing them in volume. Rather than wait, we said, ‘Hey, we know cameras work.’ ”

The company was founded in December 2016 and went through the Y Combinator business incubator for two months. Over a three-month span earlier this year, Lvl5 recruited 2500 drivers and collected 500,000 miles of driving data from them. Kouri said this accounts for coverage of 90 percent of U.S. interstates, but he added that the company also would be interested in partnering with a big automaker to help get comprehensive coverage of roads worldwide.  Lvl5 is already conducting pilot projects with several unnamed OEMs.

“We want to collect a diverse data set to see what roads look like in the snow or, for example, desert video from Abu Dhabi,” he said. “We couldn’t detect the lane lines there, and it turns out the roads were covered in sand. That messed up our algorithm, but we learned from it.”

Drivers can earn more for providing data from roads the company hasn’t yet mapped versus well-traveled roads. For ordinary commuters, earning a few cents per mile probably isn’t appealing. But it could be more enticing for drivers already working behind the wheel. For example, a trucker who drives 500 miles per day could earn, at four cents per mile, an extra $20. Do that multiple times per week, a few weeks per month. Right now, the Payver app is only available for iPhone users. Lvl5 says it is working on an Android version that should be available within a few months.

It could provide more than pocket change, and there’s little effort required. For Lvl5, the big question remains: Will enough motorists be interested?

Let's block ads! (Why?)



from Car and Driver BlogCar and Driver Blog http://ift.tt/2gQCBTf
via IFTTT

0 comments:

Post a Comment