Is it always the case that this will look badly? Or does it behave fine in certain cases?
Would we be able to get a "nice" looking transition by decomposing the start and end transformations into their respective rotations, scales, etc. and linearly interpolating those, directly?
Is there a degree of change where the transition wouldn't need additional interpolation in-between? Just like in traditional animation, with the key frames and the in-between connecting frames, how much of a change from 'frame' to 'frame' would be the limit with such a movement?
Is this just so that it looks good for a human? Or is there an objective way to know if an interpolation "looks good"?