Trajectory tracking model predictive control for mobile robot based on deep Koopman operator modeling

Trajectory tracking serves as a pivotal performance metric for mobile robot systems, and is crucial for improving the efficiency of robots. The intricate kinematic and dynamic properties of robot systems pose substantial challenges in achieving accurate modeling and effective control, which remain pressing issues within the current research domain. This study focuses on wheeled mobile robot, relying on the deep Koopman operator theory, combined with the extended state observer (ESO) and the adaptive predictive time domain self-triggered model predictive control (APST-MPC) method, to propose a data-driven solution for the trajectory tracking control issue of wheeled mobile robot under uncertain model parameters. Firstly, the dynamic model of the mobile robot is constructed by the deep Koopman operator method. Secondly, to counteract operational disturbances encountered by the robot, an ESO is designed for disturbance estimation and subsequent compensation within the controller. Thirdly, to reduce the computational load, APST-MPC is employed to enhance the trajectory tracking control of wheeled mobile robot. Ultimately, the efficacy of the proposed trajectory tracking controller is confirmed through simulation experiments. The simulation outcomes confirm the deep Koopman operator theory’s efficacy in establishing a robot model with considerable accuracy, the tracking error of the robot is reduced by 46.03% and the total number of triggering times of the system is reduced by more than 59.8% by the APST-MPC controller based on ESO compared with the MPC controller.