I live near Waymo and see the cars all the time - they do have huge things on top but the weather in Santa Monica is also very good. Not sure if I have seen them ever driving in the rain?
At the same time advancements also happen at the multi-model side, not fusing different deep learning outputs afterwards but right in the one model (not judging if that is a good idea... but the steps taking currently seem to get bigger and better).
Lastly, also camera has degraded performance under certain weather conditions.. some even say our eyes can be prone to that! :D
So that's just all the challenges that need to be taken. Yes, humans can drive with their eyes only, but is that perfect? Sometimes we e.g. would also benefit greatly from radar/lidar, as e.g. in dark conditions our judgement of distances is really flawed.
Refresh rate of some of this already aged car lidars was like 10 Hz, good enough I'd say? And it is truly amazing what details one can get out of those when getting close and move around in those point clouds.