Not only that, but it should be pretty clear that using a combination of vision technologies (cameras, LIDAR, RADAR, sonar...) means that more edge cases should

Not only that, but it should be pretty clear that using a combination of vision technologies (cameras, LIDAR, RADAR, sonar...) means that more edge cases should be covered. The failure conditions for optical wavelength cameras are different than for LiDAR, or infrared cameras, so really good sensor fusion should theoretically reduce the number of situations in which the internal model of the world has major errors. I'd rather be in a self driving car with 6 cameras and a 360-degree LiDAR then one with 20 cameras, even if 20 cameras are providing more information about the world.