Yandex makes progress in AV navigation of snowy roads

Rain, sleet and snow are among the trickiest road situations for self-driving vehicles to navigate. Not only do these conditions lead to dangerously slippery roads and hide critical lane markings, they can also reduce visibility and prevent self-driving systems from recognizing or registering obstacles that should be avoided.

Russian tech company Yandex said it has made progress in autonomously maneuvering through snowy conditions. The company recently hit 6 million miles in autonomous mode, primarily in Moscow, during a year with record snowfalls.

Wintry weather can impair the visibility of both a human driver and a self-driving system, said Yandex, which tests a fleet of about 160 vehicles across Moscow and Innopolis, Russia; Tel Aviv; and Ann Arbor, Mich. Snow and ice can also lead to condensation clouds, which could appear as solid obstacles to lidar sensors.

Companies have been working to solve vision challenges in the snow with a variety of technologies. For instance, Google subsidiary Waymo has used machine learning to help filter noise from snow out of its sensor data. Startup WaveSense uses ground-penetrating radar to "see" through rain, fog, dust and snow.

Yandex enhances its lidar technology with neural networks that filter snow from the lidar point cloud. This boosts the visibility of objects and potential obstacles, the company wrote in a blog post on Medium. By contrast, some systems' lidar beams could reflect off the snow and prevent the systems from successfully navigating through. Yandex said its vehicles generate more than 1 million beams per second.

Yandex also detects potential exhaust fumes and condensation clouds and filters this information to the system to ignore when navigating, company spokeswoman Yulia Shveyko said.

"When we filter those obstacles, which are not obstacles really — fumes, exhaust or snow — our vehicle knows that for planning its route, it can go through those obstacles because it's either fumes or exhaust or snow, and it's not dangerous at all," Shveyko told Automotive News. "That shows our technology to be highly adaptable for new locations, especially those which are easier."