Dealing with the complexities of Inverse Kinematics is essential for effectively controlling exoskeletons in complex environments. Traditional analytical methods frequently encounter difficulties due to the non-linearity of these problems. Our approach begins with using Particle Swarm Optimization (PSO), Genetic Algorithm (GA), BAT Algorithm (BAT), and Grey Wolf Optimizer (GWO) to address the inverse kinematics model for two-degree-of-freedom (2-DoF) upper Limb Exoskeletons. These metaheuristic solutions are then enhanced by integrating ANNs, utilizing both random step-size and sinusoidal-signal datasets to refine the results. This hybrid method optimizes the ANN hyperparameters, such as the number of hidden layers and activation functions. Our methodology introduces unique hyperparameter optimization techniques for training the ANNs, effectively leveraging the strengths of both metaheuristic algorithms and ANNs. Experimental results demonstrate the obtained lowest error rates, alongside the best hyperparameters for the ANN architecture. These findings suggest the scalability of our hybrid model to more complex robotic systems.