KNODEOB‐MPC: A Knowledge‐Based Neural ODE With Online Observer MPC Framework for Quadrotor

We consider the trajectory tracking control problem of a quadrotor in the presence of model uncertainty and external disturbances. Although machine learning has been effective in addressing model uncertainty through offline training, it fails to account for real‐time disturbances. Online learning, though capable of adapting to such changes, is often computationally intensive. We propose a Knowledge‐based Neural ODE with an Online Observer MPC framework (KNODEOB‐MPC) to solve this problem. Firstly, we use the neural ODE to learn the residual model of the system from the data and integrate the learned residual model with the first principles to build a hybrid model. Then, we design an online disturbance observer based on the hybrid model to compensate for the real‐time disturbances that exist in actual operation. The hybrid model and the observed disturbance information are employed in the MPC framework to realize the tracking of the reference trajectory. Furthermore, we provide a formal stability analysis of the proposed method. Simulated experiments demonstrate that the KNODEOB‐MPC framework substantially enhances the trajectory tracking performance of a quadrotor under disturbances and model uncertainty.