Title: Data-driven feedforward control design for nonlinear systems: A control-oriented system identification approach

Abstract: Feedforward controllers typically rely on accurately identified inverse models of the system dynamics to achieve high reference tracking performance. However, the impact of the (inverse) model identification error on the resulting tracking error is only analyzed a posteriori in experiments. Therefore, in this work, we develop an approach to feedforward control design that aims at minimizing the tracking error a priori. To achieve this, we show that the norm of the tracking error can be upperbounded by the sum of the inversion error and the identification error. This yields a two-step feedforward control design procedure that consists of a control-oriented system identification, and a finite-horizon optimization to compute the feedforward control signal. The nonlinear feedforward control design method is validated using physics-guided neural networks on a nonlinear, nonminimum phase mechatronic example, where it outperforms standard ILC.