An adaptive direct data-driven controller for nonlinear system reference tracking

Reference tracking, one of the primary objectives in system control theory, is increasingly being approached in a data-driven manner. In this paper, an Adaptive Trajectory Data-Enabled Predictive Controller (ATDeePC) algorithm is proposed that locally constructs a linear model of the system and stores its predictions for future use. Based on the online recorded trajectory and a desired reference trajectory, the most appropriate trajectory of the linearized model is selected for Hankel matrix construction. The proposed controller is safe in closed loop, satisfies the desired constraints during operation, and is robust with the ability to adapt to previously unexplored regions of the system state space. The controller is tested and evaluated on a simulated unmanned surface vehicle (USV) and compared with common reference tracking methods.