A hybrid method using FABRIK and custom ANN in solving inverse kinematic for generic serial robot manipulator

Solving inverse kinematic (IK) of general robot manipulators remains significant challenge in current industrial manufacturing, particularly in human–robot collaborative scenarios. Most current approaches employ numerical, analytical, or machine learning methods to solve IK. However, accurately determining the end-effector (EE) position, solving complexity, and handling multiple solutions are unresolved challenges in these existing methods. In this paper, we propose a hybrid method that combines forward and backward reaching inverse kinematics (FABRIK) with a custom artificial neural network (ANN) to solve IK for a broad range of serial robot manipulators. The results demonstrate that the hybrid method yields a unique solution and achieves a lower position error (up to 0.003 in) compared to a standard ANN implementation. Furthermore, compared to the numerical method (FABRIK and Jacobian), the hybrid approach offers a more versatile framework for solving IK, resulting in superior overall performance in terms of solving complexity, computational efficiency, and accuracy among the three methods.