Last year, in a Detroit-area parking lot, the driving seemed hopeless. Sam Abuelsamid sat in a new Kia Sedona minivan with some of the most advanced driver’s assistance features on the market, and all it would do was beep at him uselessly.
Snow was falling and the van’s rear parking assist camera was covered with ice; the ultrasonic sensors that warn of nearby objects were pinging off the snowflakes.
“We’re a lot farther from general use self-driving cars than those in Silicon Valley would like you to believe,” wrote Abuelsamid in a Facebook post at the time. “I had to turn off the parking assist because the falling snow was triggering the ultrasonic sensors causing the system to beep continuously while there was nothing around the vehicle.”
Abuelsamid knows about these things. He’s a senior automotive analyst with Navigant Research, and the Kia was a tester on loan for him to experience its features. Most of the time, he was impressed with it staying in its lane and keeping its distance from other vehicles, but when it snowed — that all became useless.
“I’m sure we’re farther ahead now, but I don’t know how much farther,” he says today. “There are always advances happening, but they’re probably closer to small steps than giant leaps.”
But yes, there are small advances for autonomous winter driving being made all the time. This year’s new Mercedes-Benz E-Class, for example, has heated radar covers that are coated with a special plastic film for protection from ice and rain. It means the radar still works in bad weather so the car can recognize guardrails and even follow other vehicles on the road through snow and storms.
This was developed after extensive winter testing in Sweden, says Michael Hafner, Mercedes’s Director of Driver Assistance Systems and Active Safety.
Normally, the car uses both its cameras and radar to follow lane markings and detect objects on the road, but “if the conditions are degrading, because of snow or rain or whatever, other objects can grow more dominant,” says Hafner. Those can be guardrails or signs or any structure, and even a changing road surface.
“We include many of these parallel structures in our (mapping) records, and the level of accuracy is high enough now that we know where the road goes, and we can keep under control on the road.”
Last month, Ford took its testing a step farther after a snowfall at the University of Michigan’s Mcity proving ground. The 13-hectare facility has a network of roads and intersections set into a fake town for safely testing vehicles.
“One of the real challenges is to understand the things that we don’t understand. Going into snow falls into that,” says Jim McBride, Ford’s Technical Leader of Autonomous Vehicles.
“We were looking for an opportunity to have snow fall on the ground and go out and drive, and say look, we don’t see the painted lines, we don’t see the crosswalks, we don’t see any of that, but we see the rest of the world around us, and that’s really what enables the vehicle to know where it’s located.
“There were no surprises whatsoever — the vehicle did exactly what it’s supposed to do.”
The Ford Fusion Hybrid was considerably more advanced than any current production vehicle, however. For one thing, it’s equipped with spinning Lidar sensors on the roof, which emit laser light for up to 65 metres that creates a highly accurate computerized 3D image from the reflections.
This light bounces off raindrops and snowflakes, but the computer is smart enough to recognize them as such and does not let them affect the image.
The test car also reads from a mapping program so detailed and accurate that it can place the vehicle within a centimetre of its location at any moment. Normal GPS is only good for a metre or two. Like a blind person walking through a familiar room, it knows exactly where everything is; should anything change, the Lidar will inform it.
Google’s driverless car operates in a similar way, and the company says it’s now testing it in “rainy and snowy conditions,” but has not reported on its progress.
McBride is optimistic these “enabling technologies” are close to production, perhaps within five years. So is Hafner, who acknowledges 2020 may be “the magic year.” Not for totally driverless cars, perhaps, but for much more autonomy than now.
Abuelsamid is not so sure. “If your Lidar sensor is covered with road salt, it doesn’t matter how good it is,” he says. “Autonomy’s not going to happen overnight — it’s going to take a while.”
0 Comments:
Post a Comment