A novel algorithm by combining nonlinear workspace partition with neural networks for solving the inverse kinematics problem of redundant manipulators

Redundant manipulators (RMs) have been gaining more attention thanks to their excellent merits of operating flexibility and precision. Inverse kinematics (IK) study is critical to the design, trajectory planning, and control of RMs, while it is usually more complicated to solve IK problems which may inherently have innumerable solutions. In this work, a novel approach for solving the IK problems for RMs while retaining the redundancy characteristics has been proposed. By employing a constraint function, the method delicately reduces the infinite IK solutions of a RM to a finite set. Furthermore, the workspace of RMs is divided into nonlinear partitions through diverse joint angle intervals, which have further simplified the mapping correlations between the desired point and manipulators' joint angles. For each partition, a pre-trained neural network (NN) model is established to acquire its IK solutions with high efficiency and precision. After combing all nonlinear partitions, multiple reasonable IK solutions are available. The presented method offers a possible selection of the most appropriate solution for trajectory planning and energy consumption and therefore has the potential for facilitating novel robot development.