Ground-penetrating radar may soon be the sensor that makes your car autonomous in all weather conditions. It turns out that when you scan the 10 feet below the roadway surface, you get a unique identifier that is accurate to an inch or two. Mapping cars would scan the roadways once, then your self-driving car with its own ground-penetrating radar would rescan as you drive, matching its real-time scan to the master map. That would keep your car centered, even if pavement markings are covered by snow or ice, according to WaveSense, an MIT spinoff that already has already tested military applications.
Ground-penetrating radar can’t be the only sensor in a self-driving car. An autonomous car still needs surface radar, possibly lidar, and cameras to track other vehicles, pedestrians, animals, blocked lanes, and cars stopped or crashed in travel lanes. But it has the potential to be the breakthrough that allows bad-weather autonomous driving.
To the human eye, every road looks about the same, give or take the number of potholes, and how much the lane markings have faded. But the subsurface combination of rocks, cavities, culvert pipes, utility infrastructure (cables, conduits, sewer lines), and reinforcing steel bar for concrete (rebar) creates a radar image uniquely different from any other part of the roadway.
Each initial mapping pass covers almost the width of a highway lane. WaveSense president and co-founder Tarik Bolat says the map WaveSense creates is “rich in detail, stable, and always available.” The accuracy at highway speeds can be as good as 4-6 cm or 1.4-2.4 inches depending on weather, he says. A passenger car is about six feet wide, while a truck is eight feet wide and a highway lane is 12 feet wide. So the location accuracy is 25 times better than it has to be to stay in lane.
Tested First by the Military
Byron Stanley, WaveSense’s co-founder and chief technology officer, says the concept evolved at MIT’s Lincoln Laboratory at Hanscom Air Force Base in Lexington, Massachusetts. The Lincoln Lab creates technology for national defense.
The GPR concept dates to 2013 and a project to make 6×6 military trucks safer in war zones by potentially swapping the driver for an autonomous system. Such trucks would operate on roads with no lane or edge markings, no road signs (or signs deliberately switched), and harassing fire that might affect a driver’s concentration.
What did the tests find? “There was a real possibility of a significant impact on the safety metrics of existing autonomous vehicle fleets,” Stanley said. It became clear there was a civilian market awaiting as hardware costs came down for ground-penetrating radar as well as for the sensors auto- and truck-makers would need as well. WaveSense was founded in 2017. As is customary, MIT receives future proceeds for having been the incubator.
According to Bolat, ground-penetrating radar could be on autonomous cars circa 2024 with a cost to produce on the order of $100 in quantity. WaveSense believes lidar may not be necessary; lidar is currently the most expensive sensor system on prototype cars. Lidar provides a high-resolution map of what’s around a vehicle, although the range and image are reduced in snow or rain.
A WaveSense scanned public parking garage in Madrid. Concrete, rebar (steel reinforcing bars), and conduits create a signature just as on public roads.
Works in Parking Garages, Too
Interestingly, WaveSense believes the use of GPR could extend off public highways. It might enable the automated parking garage. The arrangement of concrete, rebar (reinforcing metal bar), and conduit creates the same kind of unique digital signature in a parking deck, underground garage, or large surface lot. Just as on the highway, there would have to be cameras, radar, or (possibly overkill in this instance) lidar to detect garage walls, posts, and other cars. The garage structures would also have to be pre-mapped, just as on highways.
Pre-mapping is also the technology that has lifted General Motors’ Super Cruise above all other current self-driving technologies. Before GM launched Super Cruise on its Cadillac brand in 2017, it sent mapping cars to lidar-scan all interstates and similar roads in the US and Canada. That gave GM a map of roads, road edges, bridges, and other obstacles near the road. That exact location information then goes into production cars to assist the existing sensors – radar, cameras – determine the car’s exact location.
Unlike WaveSense that calls for GPR for pre-mapping and in end-user vehicles, Super Cruise needs lidar pre-mapping but doesn’t require lidar in production cars. It relies on cameras and radar to locate the car relative to lane markings, along with the lidar map that links to GPS location info. GM recently announced it would expand Super Cruise to about 20 car models across the GM lineup, which should further reduce the cost, perhaps to around $2,000. In past Cadillacs, it was part of a $5,000 options package.
Now read:
- TriEye’s Infrared Camera Helps Autonomous Cars See Through Haze
- The Future of Sensors for Self-Driving Cars: All Roads, All Conditions
- Cadillac Super Cruise Adds Auto Lane Change
from ExtremeTechExtremeTech https://ift.tt/2HzStEn
ليست هناك تعليقات:
إرسال تعليق