This paper describes the learning method to create nonlinear transformation of inverse kinematics using a three-layer neurall network, showing effects of network parameters such as output function, number of hidden units, learning rate, momentum constant through simulations.
