NaNs appear in certain cases when using inverse kinematics

[64-B1T]

(Python) In certain transforms, IK will fail due to NANs cropping up. This is because of floating point inaccuracies being fed into np.arccos. (For example in MatrixLog6)
E.g. (-1.0000000002) which will return NaN
To fix this, for each np.arccos(x) in the MR code, replace with np.arccos(np.clip(x,-1,1))

[HuanWeng]

Thank you for pointing this out. Looks like you've fixed that in your branch. I will check and fix that in our next update.