Model-Free Derivative Feedback Control for Active Magnetic Levitation System

We present a model-free, data-driven state-derivative feedback controller for active magnetic levitation. An online policy-iteration scheme updates the control law from real-time state and input data, directly minimizing a quadratic cost while guaranteeing closed-loop stability. Training alternates between data collection and policy updates over several epochs to reduce bias, and the best policy is selected at the end. Hardware tests confirm robust performance despite dynamic and equilibrium uncertainties.