Forward Kinematics of Delta Manipulator by Novel Hybrid Neural Network

Abstract

For the parallel configuration of the robot manipulator, the solution of Forward Kinematics (FK) is tough as compared to Inverse Kinematics (IK). This work presents a novel hybrid method of optimizing an Artificial Neural Network (ANN) specifically Multilayer Perceptron (MLP) with Genetic Algorithm (GA) and Step-wise Linear Regression (SWLR) to solve the complex FK of Delta Parallel Manipulator (DPM). The joint space angular positional data has been iterated using IK to generate point cloud of Cartesian space positional data. This data set is highly random and broad which leads to higher-order nonlinearity. Hence, normalization of the dataset has been done to avoid outliers from the dataset and to achieve better performance. The developed ANN based MLP gave a mean square error of 0.0000762 and an overall R2 value of 0.99918. Finally, the proposed network has been simulated to solve FK of the parallel manipulator and to check its efficacy. For given joint angles, the proposed network predicted positional values which are in good approximation with known trajectory solved by standard analytical method.