Optimized Inverse Kinematics of a 2-DoF Robotic Manipulator Using a Hybrid Approach Combining an ANN with a Metaheuristic Algorithm

Abstract

Resolving the Inverse Kinematics (IK) problem is critical for investigating 2-DoF robotic manipulators in complex environments. Traditional analytical methods often struggle with the inherent nonlinearity of these problems. This paper presents a hybrid approach combining Particle Swarm Optimization (PSO) with Artificial Neural Networks (ANNs) and Genetic Algorithm (GA) with ANNs to to obtain the hyperparameters for the architecture of ANNs such as the hidden layer size, and the activation function. The proposed methodology introduces a hyperparameter optimization technique for training the ANNs, using a Random step-size dataset, effectively leveraging the strengths of both metaheuristic algorithms and ANNs. Experimental results are conducted to show the efficiency of the proposed hybrid approach in the computation of the IK of the 2-DoF robotic manipulator, with the Mean Squared Error (MSE) values for PSO-ANN and GA-ANN being around 8.7954e-7 and 8.9492e-5, respectively.