Data- und Learning-Based Control

Data- and Learning-Based Control

Block diagram with dlc structure Block diagram with dlc structure Block diagram with dlc structure © IRT

In many practical control applications, the mathematical model of the system to be controlled or the environment of the system is unknown or uncertain, making the use of most standard control designs impossible. Data-based control is the area of research that investigates the development of stabilizing controllers using only suitable collected input-output information, without any knowledge of the system model. Similarly, learning-based control uses artificial intelligence or online learning algorithms to learn the model or the environment of the system or directly determine a stabilizing controller from input-output data. These control strategies have strong potential applications for control of nonlinear systems, where obtaining accurate system models is often difficult or practically unfeasible.

In particular, our current research focuses on

  • online convex optimization for control of dynamical systems,
  • data-based descriptions of nonlinear systems,
  • the development of data-driven predictive controllers,
  • the design of off-policy reinforcement learning algorithms to obtain state-feedback controllers, and
  • the study of data-based controllers for irregularly sampled systems.