Solving Inverse Kinematics of a 2-DoF Upper-Limb Exoskeleton Robot Through Metaheuristic Optimization of ANN Hyperparameters

For two-degree-of-freedom (2-DoF) exoskeleton robots, achieving efficient and accurate solutions for Inverse Kinematics (IK) is critical for precise robotic movements and tasks during rehabilitation. Traditional analytical methods for solving these IK problems can be complex and computationally expensive, often struggling with non-linearity issues and multiple solutions. To address these challenges, this study explores the utilization of metaheuristic optimization algorithms (MHOAs), specifically the Dragonfly Algorithm (DA) and the Multi-objective Dragonfly Algorithm (MODA), combined with an Artificial Neural Network (ANN) to develop intelligent and efficient approaches for computing IK solutions of a 2-DoF exoskeleton robot. These MHOAs are used to find the optimal hyperparameters of the ANN, which is composed of one hidden layer. Extensive simulations and experiments are conducted on the exoskeleton to evaluate the performance of the ANN-DA and ANN-MODA hybrid models. The results demonstrate that the fitness values obtained are very low, approximately equal to zero, indicating precise predictions by the models. This highlights the ideal convergence between the desired and estimated end-effector positions. The high level of accuracy validates that the MHOA-based ANN models are well-trained and capable of providing reliable solutions for the IK of the proposed exoskeleton robot.