Synthesis of Real-Time Control Systems for Multilink Industrial Robots Based on Hybrid Neural Network Approach of Solution Inverse Kinematics Problem

Abstract

The paper considers the issue of constructing real-time control systems based on a new developed hybrid method for solving the inverse kinematics problem. A method based on neuro-fuzzy network (ANFIS) with subsequent iterative refinement of the obtained solution by Newton-Raphson numerical method is proposed. Studies of the applicability of this method in control systems for three-, five-and eight-link robotic manipulators structures are carried out. The comparative analysis of methods for solution of IK problem: classical iterative, method based on neural network and the proposed hybrid is presented. The possibility of adaptation of neural networks for the developed IK method using a corrective buffer is considered. The solution as the system of equations for direct kinematics of multi-link structures is given.