Design of Event-Triggered Controllers Using Gaussian Processes

Abstract

In networked control systems, it is an important issue to reduce communication costs using aperiodic communication. In this paper, we propose a method to design the event-triggered controllers, which contains the control policy and the communication policy, for uncertain discrete-time nonlinear systems with disturbances. The control policy determines the control inputs to drive the system to the target state, and that the communication policy determines whether the control inputs should be transmitted or not based on the communication cost and the control performance. In the proposed method, two Gaussian Process models represent the uncertainty in dynamics and a state-valued function to design controllers using policy iteration. We demonstrate the numerical example to show how our learning algorithm works.