How practical would it be to include LIDAR in the initial real-world environmental scan to get (or at least seed/constrain with real data samples) an even better collision mesh?