@youen one non-obvious thing that might work: precompute a set of quaternions that you want to allow (eg. By a set of quantized euler angles). Then, you snap to the closest of those quaternions (cos theta distance might suffice). That swaps the problem around: you always have a strong guarantee of a small discrete set of possible poses, but now the difficult part is choosing a good set of poses.PS: I think order independence will never work out. It's not even true for 90° I think.