A modular neural network architecture for inverse kinematics model learning

In order to reach an object, we need to solve the inverse kinematics problem, i.e., the coordinate transformation from the visual coordinates to the joint angle vector of the arm. The learning of the inverse kinematics model for calculating every joint angle that would result in a specific hand position is important. However, the inverse kinematics function of the human arm is a multi-valued and discontinuous function. It is difficult for a well-known continuous neural network to approximate such a function. In order to overcome the discontinuity of the inverse kinematics function, a novel modular neural network architecture is proposed in this paper. (C) 2001 Published by Elsevier Science B.V.