This work summarizes a line of research focused on advancing model-free and data-driven approaches for the control of robotic systems. Traditional model-based control strategies have proven effective in structured settings, but their applicability is hindered when facing strong uncertainties, unmodeled dynamics, and structural changes. Our contributions span three main directions: (i) a novel framework for model-free kinematic control applicable to both cable-driven parallel robots and tendon-driven soft robots, ensuring convergence without learning phases; (ii) a position-based visual servoing scheme that eliminates the need for hand–eye calibration, achieving robust target tracking with only approximate knowledge of camera-to-robot pose; and (iii) a data-driven relatively optimal control method for linear systems, enabling closed-loop controllers to be designed entirely from experimental trajectories. These contributions demonstrate how model-free plant tuning and data-driven techniques can enhance robustness, flexibility, and practical deployability of robotic controllers in uncertain and dynamic environments.
