How would satellite imagery help at night, during cloudy/stormy weather? What sort of latency does the imagery have? Can commercial imagery even resolve things like painted lane markers, construction detour signage, etc?
Part of the problem is reacting in real-time. The part I'm talking about: Is you can "tell" the car what to expect at this interchange, and still react in real-time to present conditions. You hear all the time about Teslas driving off a ramp or not changing lanes quick enough due to odd paint stripes on the road, and I keep thinking you can see this from space. Why isn't there a database of known "this is a road, this is not a road" that it looks at first before looking at surrounding cameras outside the car? You can see how traffic moves -> from space. Part of these conditions are known well ahead of time on any present day regardless of the weather.
There is such a database, but Tesla relies on camera images for the data while nearly everyone else uses LIDAR to create a 3D model in addition to camera images (which can be mapped onto the 3D point cloud, enhancing their value). Camera-only is a bet-the-company move considering the large slice of Tesla'a valuation based on future robotaxi revenue.
There’s also a much easier method - just get the delta of every Tesla on the road to the speed limit and you can see how well traffic flows. All the navigation apps already do that
car meet deer meet death in my neck of the woods anyway. Satteltes can't predict deer. Most of the time we can't even predict them.