Re: the above truck picture, we've all been there in the northern states. Not like Kentucky.
There's a textbook approach autonomy uses called Simultaneous Localization and Mapping (SLAM), where perception data is compared with previously generated models and live data to perform both localization (where are we?) and mapping (what's around us?) simultaneously.
In the snow example it helps because if we've driven on that road before we can perform localization from terrain features. Too bad there are none in that environment, and the ones that do exist (fence posts) can get confusing since they match a lot of different areas. But snow ruins it a lot because it changes the terrain in unpredictable and system error-causing ways.
All that to say, yeah, snowfall is tough.
There's a textbook approach autonomy uses called Simultaneous Localization and Mapping (SLAM), where perception data is compared with previously generated models and live data to perform both localization (where are we?) and mapping (what's around us?) simultaneously.
In the snow example it helps because if we've driven on that road before we can perform localization from terrain features. Too bad there are none in that environment, and the ones that do exist (fence posts) can get confusing since they match a lot of different areas. But snow ruins it a lot because it changes the terrain in unpredictable and system error-causing ways.
All that to say, yeah, snowfall is tough.