> all of your sensor systems need to notice a danger...before you can react
I'd be really surprised if they implemented it this way. I thought the whole pitch behind lidar was that if vision doesn't see the person walking in front of you, lidar can still see the moving person shaped thing and you can still act.
If your safety reaction is consequence-free, no problem. If it isn't (like slamming on brakes in traffic), you really want a 3rd sensor so you can do 2oo3 voting.
Suppose you have a color sensor and a form sensor and you have to identify an orange (fruit). Clearly in conjuction the two sensors will be way better, than assigning everything with color orange orange and everything round an orage.
If the signal goes from "black nothing" to "orange round" you know you went from no object to an orange, but what if your form sensor breaks and it goes from "black nothing" to "orange nothing?"
Sensor failures are independent events and need to be correctly identified no matter what.
To take Lidar and Vision, both say something about distance and form of objects. Together they can achieve better performance. If one of them fails (and failure is identified) it defaults to the other sensor only and will be somewhat worse, but should at least allow it to pull over and warn you about sensor failure.
Also failures are much easier to identify when you have a baseline reference of one or more other sensors, in isolation much harder.