Neural networks based real time solution for forward kinematics of a 6 × 6 UPU flight simulator

The advanced capabilities of parallel robots have led to a dramatic increase in their use over recent years. As a result of this development, the search for solutions to the forward kinematic problem in real-time applications has gained in prominence. The lack of analytical solutions for this problem makes numerical methods and artificial neural network-based approximators popular methods for real time applications. In this paper, three distinct neural network-based approaches—namely, Multilayer Perceptron, Radial Basis Function, and Adaptive Neuro Fuzzy Inference System (ANFIS)—are designed to solve the forward kinematic problem (FKP) of a 6 × 6 UPU (universal-prismatic-universal) mechanism in real time. The developed approximators are implemented into a 6 × 6 UPU Stewart Platform-based flight simulator and the performance and computational time of the typical Newton Raphson method and three designed neural network-based structures are compared. The results of the study indicate that neural network approaches are appropriate alternatives to the classical iterative NR method in terms of real-time solution to the parallel mechanism Forward Kinematics (FK) problem. Although the MLP method has slightly better results than the RBF method in terms of computational time, the RBF method is the best in positioning accuracy.