General Motors‘ Cruise subsidiary is making huge progress on its autonomous robotaxi project, but the company has not yet demonstrated how its autonomous technology would work on a road where the lane markings are covered by a layer of snow. Other autonomous vehicle companies, such as Google’s Waymo and the Ford-backed Argo AI, are actively testing their vehicles in winter conditions, but media demos and other public demonstrations of AV prototypes are almost exclusive to warm, sunny places like Arizona, Nevada and California.
While most of today’s AV prototypes would evidently struggle in snow, researchers at MIT’s Computer Science and Artificial Intelligence Lab (CSAIL) are working on ways to improve AV performance when the lane markings may not be visible. Using a specialized groud radar, referred to as ground-penetrating radar (LGPR), MIT is able to map the soil layers, rocks and the road bedding to localize the vehicle to “centimeter-level accuracy.” Put more simply, the car is able to lane-keep by studying the layout of the ground beneath the road, rather than the painted lane markers on the surface. As more LGPR-equipped vehicles drive in an area, the accuracy of the subsurface data becomes more accurate, helping the system accuracy to improve over time.
LGPR technology was developed at the MIT Lincoln Laboratory. The school reached an agreement with Geophysical Survey Systems, Inc. (GSSI) to develop commercial prototypes of using the technology in 2017, which it is now putting to use in real-world prototypes. Byron Stanley, MIT’s lead researcher for the LGPR program, said LGPR “could significantly impact the self-driving vehicle industry,” by allowing AVs to map data on both the road surface and the subsurface.
“Even in fair conditions, having an independent sensor to rely on when your optics aren’t working could add several orders of magnitude to the reliability of current autonomous lane-keeping systems,” he said.
This technology could also potentially help when an AVs sensors and cameras are covered in ice and snow, or even if forward visibility is bad in a blizzard, rain, or dust storm. Inclement weather still presents some additional challenges for AV engineers, though, as inclement weather conditions may cause slippery roads and cause other motorists to drive unpredictably.
Watch as MIT demonstrates its LGPR-equipped AV on a snow-covered road in the video embedded below.