Event-triggered control with reliable Gaussian process learning for remote UAV control

Abstract

An event-triggered control strategy executes control updates only when a system needs attention. This strategy has been validated as effective in networked control systems by reducing data communication load. To maximize the efficiency of an event-triggered controller, disturbance compensation is essential. In this study, we employ Gaussian Process (GP) learning to estimate model uncertainties. The GP model is trained and updated online using streaming sensor data, with sparse approximation techniques applied to ensure computational tractability and real-time inference without compromising control responsiveness. The key contribution of the proposed event-triggered controller with GP learning is its guaranteed stability, achieved through an analytical error-bound inequality. This stability ensures a reliable operational range for the control system, enabling secure and adaptive adjustment of event-triggering parameters. Applied to a networked quadrotor flight control system under wind disturbances, the proposed method demonstrates accurate and efficient control performance while remaining computationally feasible for real-time implementation.