At present, it's _not_ clear that lidar would solve their issues, because from the FSD beta footage that we've seen it doesn't look like the problem is with _perception_. Perception-wise, the car seems pretty solid, but its ability to shoehorn a given world state into something it knows how to react to seems shaky. We can see examples of this where the car gets confused at funky intersections (e.g., multiple sets of lines that look like crosswalks) or challenging left turns (it can resolve cars with enough fidelity it seems, but is missing a policy for turning into the median first).

If it were as simple as "their system would just work if there was lidar" I'd bet there'd be a lidar test fleet of FSD teslas that would train the lidar-blind cars in a sort of shadow-mode, similar to what was done for vision-only replacement of radar.