I would assume that you would have this data about every object that the system can recognize. It would just be trained like any other system would be trained, because you can use the ladar for finding out the real distance. It, of course, will be difficult to guess the distance on objects the main system haven't been trained for.
The interesting part is that distance is not really needed, only good estimate of presence or absence of an object and a real easy model of how a mirror works.
