Hybrid Metaheuristic and Artificial Neural Network Approach for Solving Inverse Kinematics of a SCARA Manipulator Robot

Abstract

This paper presents a hybrid approach that inte-grates metaheuristic algorithms and Artificial Neural Networks (ANNs) to address the Inverse Kinematics (IK) problem of a SCARA (Selective Compliant Assembly Robot Arm) manipulator robot with four degrees of freedom. The method combines Particle Swarm Optimization (PSO) with ANNs and Genetic Algorithm (GA) with ANNs to optimize training key hyperpa-rameters, such as activation functions and hidden layer sizes using MATLAB's Neural Network Toolbox. Experimental results achieved on a random step size dataset obtained after training show that the PSO-ANN method achieves a Mean Squared Error (MSE) of 0.12846, with a hidden layer size of 90. The GA-ANN method results in an MSE of 0.13785, with a hidden layer size of 91. This hybrid approach significantly reduces MSE in computed joint configurations and demonstrates promise for real-time control applications.