Event-triggered anytime control with limited resources

Abstract

In networked and multi-tasking environment, measurement data and processing resources may not be available at times when control calculations need to be executed. Based on the anytime algorithm been proposed in Stochastic Stability of Event-triggered Anytime Control[1]for control of event-triggered systems(nonlinear)which the processing resources available are time-varying, the algorithm recursively calculates a sequence of tentative plant inputs when a plant state measurement is successfully received and while the processor is available for control, which are stored in a buffer for potential future use. This safeguards for the time-steps when processor is unavailable for control. To make more efficient use of communication and processing resources, we extend this algorithm with two controllers in this dissertation. We present an anytime algorithm which features two control policies: a coarse policy and a fine policy. The fine control policy requires more processing resources than the coarse policy. With this scheme, the network and processing resources can be used more efficiently, and performance can be improved. Specifically, for a given packet dropout rate and process availability, the proposed two-controller scheme achieves better closed-loop performance with a lower channel utilization than alternative control formulations. The stability region is also enlarged with the two-controller scheme.