Exploring Neural Networks for Forward Kinematics of the Robotic Arm with Different Length Configurations: A Comparative Analysis

Abstract

Our work investigates the utilization of Artificial Neural Networks (ANNs) to address the complexities associated with Forward Kinematics (FK) problems within the field of robotics. We undertake an extensive comparative analysis to assess how ANNs perform under different circumstances with robotic arms of varying lengths and impact the overall system’s functionality. The training, testing, and validation of ANNs are carried out using MATLAB for a simulated 2-DoF serial robotic arm involving three distinct datasets: fixed step size, random step size, and sinusoidal step size. Three training optimizers, namely Levenberg Marquardt (LM), Bayesian Regularization (BR), and Stochastic Conjugate Gradient (SCG), are considered within the ANN architecture. Based on Mean Square Error (MSE) values, the numerical findings reveal the potential of ANN in estimating forward kinematic solutions of complex robotic manipulators with different arm lengths and reducing computational complexity.