From the June 2016 issue
The cameras and radar sensors that enable today’s highly automated cruise-control systems can see only 250 yards ahead in the best conditions. That’s barely more than six seconds of lead time at 80 mph, and traffic, weather, and topography often shrink visibility to even shorter distances. For true autonomy, self-driving cars will need a better understanding of the bigger picture.
Like TomTom, Here creates its high-def maps (above) from billions of data points captured by laser Scanners (top of page).
Steady Gigs
To create HD maps and RoadDNA imagery, TomTom’s mobile mapping vans record 22 billion pixels per mile with onboard cameras, while a laser-emitting lidar scanner collects more than 700,000 data points per second. By compressing the three-dimensional data into two-dimensional images, TomTom can condense every interstate in America into roughly 20 gigabytes of data.
The coming generation of so-called high-definition maps will supply that picture, along with enough detail for computers to make predictive decisions rather than reactive adjustments. While today’s navigation maps represent each road as a single lane, HD maps include multiple lanes with usage rules, curbs, shoulders, road signs, and guardrails. Armed with high-definition maps, the smartest cars will leapfrog slow-moving traffic, negotiate interchanges, move over for merging cars, choose the correct lane for an exit, and know where to stop in an emergency—all without driver input.
In charting the course for autonomous cars, mapmakers face a major paradox: The more detailed the map, the more likely it is to be inaccurate. Lanes become shoulders, speed limits change, and construction turns meticulously planned traffic patterns into chaos. That’s why Here—a mapping company owned by BMW, Daimler, and the Volkswagen Group—envisions a future in which autonomous vehicles redraw the maps as they travel the roads.
“We map areas yearly. That’s probably not fresh enough to maintain what we need going forward,” says John Ristevski, vice president of reality capture and processing at Here. “We need to reflect any changes in the environment in real time.” A dynamic layer in Here’s HD maps aggregates live data from fixed sensors, government agencies, and other cars on the road to adapt to temporary lane closures, accidents, and traffic jams, along with permanent changes to traffic patterns.
For the mapmakers, the route ahead is marked and clear of obstacles. Here’s Ristevski predicts that the most advanced cars will drive entrance-to-exit on our highways without driver intervention or supervision within three to five years.
CENTERED
To Know Where You’re Going, You’ve Got to Know Where You Are
Even the most detailed map is useless if you don’t know your location. Consumer-grade GPS can only place a car within a 10-foot circle, so mapmakers have developed methods to pinpoint its exact position on their HD maps, a process known as localization.
TomTom augments its high-def maps with RoadDNA, which compares live data captured by an onboard lidar sensor with images previously recorded by its mapping vans. By identifying patterns in the images and subtle differences between the locations of fixed landmarks, RoadDNA can determine a car’s side-to-side position on a road to within six inches. In a typical 138-inch–wide lane, that’s precise enough to let the car steer. “A six-inch drift in one direction or the other is something that most of us can hardly perceive, even if we’re driving down the road ourselves,” says Nhai Cao, TomTom’s global product manager. Still, no one is proposing eliminating today’s camera-based lane-keeping systems. Well, not yet.
from Car and Driver Blog http://ift.tt/1WdL40j
via IFTTT
0 comments:
Post a Comment