So there's three main issues that snow causes for autonomous vehicles: - It (partially) covers lane boundaries - Phantom objects - Black snow The easiest way for autonomous cars to find their place on the road is to find the lane boundaries and use them to center the car. Snow-covered roads make that impossible. Even cleared roads often still have snowbanks on the sides, which produces noisier data. As a result, the car sometimes detects objects that aren't there, called phantom objects: Lastly, there's black snow which is really hard to detect because it's so transparent. Solutions might be V2V communications (platooning, one car leads the way and the rest follow) but it needs to be very reliable. Slowing down won't help on snow-covered roads, but it might help on (partially) cleared roads. You're right though, it's a problem that will most likely be avoided as long as financially makes sense.
Figure 12. Example of pitching combined with small pose estimation errors: a The reading of the center beam of one of the lasers, integrated over time some of the terrain is scanned twice.; b shows 3D point cloud; c resulting map without probabilistic analysis, and d map with probabilistic analysis. The map shown in c possesses a phantom obstacle, large enough to force the vehicle off the road.