From what I understand the inverse matrix is always applied but the
object-intersect functions still take arguments like position or
location, wouldn't it be better to always intersect with for example a
unit sphere and calculate the transformation matrix from things the user
specifies, i.e. location, radius, etc.. ?