On Tuesday, March 10, 2018 Patrik Berner (Ruhr-Universität Bochum, Germany) gave a lecture on "Event-Triggered Networked Model Predictive Control".
Speaker: Patrik Berner, Automatic Control and Systems Theory, Ruhr-Universität Bochum, Germany
Title: Event-Triggered Networked Model Predictive Control
Date, time, and place: April 10, 2018, 10:45am in room no. 641
Embedded hardware often does not provide sufficient computational power to compute complex optimal control problems periodically as needed for MPC. We introduce a networked approach that uses a computationally powerful central node that provides one or more lean local nodes with regionally optimal affine feedback laws. Whenever the region of optimality of a current affine control law is left on one of the local nodes, this local node requests a new optimal law from the central node. Since evaluating an affine control law and checking if the current system state lies in the region of optimality only requires simple calculations, these operations can be carried out on the local nodes.
We deduce four variants of this simple networked MPC approach and compare the computational effort and transmitted data. We show that there exists a best trade-off between transmitted data and local calculations and propose an upper bound for the number of local calculations for this particular variant. In order to reduce the computational requirements even further, we propose an approach that reduces all quantities (computational effort and transmitted data) of the networked controller by overclocking the local nodes. Theoretical results are corroborated by a hardware-in-the-loop setup that uses a microcontroller and wireless network as the communication medium.