Model-Based Control Using Data-Driven Models


Authors: Gerardo de la Torre, Ian Abraham, Todd Murphey

This paper explores the application of Koopman operator theory to the control of a robotic system. We illustrate how the approximate Koopman operator can be used to obtain a linearizable data-driven model of an unknown dynamical system. Finally, we derive closed-loop and open-loop controllers using the proposed data-driven model. Furthermore, we explore the utility of the approximate Koopman operator in a reduced state representation setting. Experimental studies using the SpheroSPRK robot motivate our discussions.