Javi has pointed out that the current function to get distance between two rotations isn't very nicely behaved in that it doesn't get close to 0 until the rotations are really, really close to one another. Going back to the simpler formulation of distance as simply the euclidean norm between the two quaternions (properly signed) divided by the sqrt(2) (to keep it between 0 and 1) should improve things. Any comment/objections?