Learning-Based Nonlinear MPC With Integrated Moving Horizon Estimation of Quadrotors

This article studies a learning-based cascade nonlinear model predictive control (NMPC) problem for quadrotor trajectory tracking under unknown time-varying disturbances. A nonlinear moving horizon estimation (NMHE) method is employed to estimate the quadrotor’s states and key parameters such as aerodynamic coefficients, mass, and moment of inertia. Based on the estimates, a differentiable NMPC, along with an online learning method to adjust the cost function considering the tracking trajectory, is developed. This enables the framework to adapt to dynamic environments. Finally, comparative studies validate the effectiveness and advantages of the proposed method.