Learning-Based Quadrotor Tracking Control Using Recursive Gaussian Processes

High-precision UAV tracking is hindered by hard-to-model aerodynamic effects. We propose a learning-based controller for a simplified quadrotor that uses Feedback Linearization (FL), to cancel modeled dynamics, and a sparse Bayesian inference method called Recursive Gaussian Process (RGP), to learn and compensate residual nonlinearities in real time. Simulations and experiments conducted on a real 1D quadrotor platform demonstrate reduced tracking error and practical feasibility for real-world implementation.