But if I have to put it in a nutshell, some of the information you need for driving (traffic signs, road semantics etc.) is only visible to the camera. Other sensors will not help you. So any benefit you may get in low visibility scenarios is useless. If visibility is low, you need to slow down, like a human. If your computer vision is good enough to read traffic signs, road semantics, and lanes, etc., then it will also be good enough to detect objects. Using sensor fusion (LiDAR, RADAR) to enhance object detection is just a cover for poor computer vision models. I know it very well, that's my job. The computer vision fails in some situations, although the object is clearly visible in the image space, and we use other sensors to "fix it". But then your computer vision will also fail at the other tasks where there is no possible redundancy. This is very visible in Rivian's latest "hype" videos. A pedestrian crossing in close range is missed for several frames by the computer vision, and Rivian "demonstrates" that their fusion probably tracks the pedestrian continuously, although the camera model has failed. To me, this video does not prove fusion is the answer; it just proves their computer vision is shit. The camera also fails to detect a bus. Pathetic when you think they present this video as "promotion". Then there are also many other arguments. Sensor redundancy introduces complexity. Active sensors are not in sync with the cameras (all cameras can capture their frames simultaneously, but this is not possible for active sensors which would interfere with each other), and it makes fusing this information more complex. Increase complexity, increase cost... more problems... I have written a detailed article on what is necessary to make FSD a reality if you want to know more.