I know at least 3 different approaches to solve inverse kinematics problem. They are pseudo inverse jacobian, cyclic coordinate descent and ANFIS networks. I would like to know advantages and disadvantages of these methods comparing to each other.
I know that CCD suffers from local minima but as I heard this problem seldom arises in practice. However it is more intuitive and easy to implement comparing to pseudo inverse jacobian. Therefore I wonder why would one use inverse jacobian instead of CCD?
While these two methods I have found to be the most popular there is another approach called ANFIS networks which requires gathering training data from forward kinematics equations and training a neural network. It seems a bit overcomplicated to me. Why would one prefer ANFIS to CCD or pseudo inverse jacobian methods?