State predictor-based deep model reference adaptive control for quadrotor trajectory tracking

The application of quadrotor unmanned aerial vehicle (UAV) swarms has attracted considerable attention in recent years, but the dense formations also pose new challenges to controlling quadrotors. In these cases, quadrotors frequently encounter matched and unmatched disturbances from fellow swarm members. To achieve precise tracking of the desired trajectory with optimal accuracy, a state predictor-based deep model reference adaptive control (PDMRAC) framework is proposed. Owing to the powerful feature extraction capability of deep neural networks (DNN) and the enhancement of transient characteristics of the system by the state predictor, the control framework's performance in approximating unstructured uncertainty is improved. The controller designed based on the nonlinear model compensates for the matched uncertainty and reacts to the unmatched uncertainty to reduce the tracking error. Moreover, the controller maintains accurate tracking performance for unseen trajectories and uncertainties when well-trained DNNs are employed as frozen weight networks. Numerical simulations are conducted to evaluate trajectory tracking performance in an environment featuring time-varying disturbances, and the results demonstrate the effectiveness of the proposed method.