This article presents new results to control process modeled through linear large-scale systems. Numerical methods are widely used to model physical systems, and the finite-element method is one of the most common methods. However, for this method to be precise, it requires a precise spatial mesh of the process. Large-scale dynamical systems arise from this spatial discretization. We propose a methodology to design an observer-based output feedback controller. First, a model reduction step is used to get a system of acceptable dimension. Based on this low-order system, two linear matrix inequality problems provide us, respectively, with the observer and controller gains. In both the cases, model and reduction errors are taken into account in the computations. This provides robustness with respect to the reduction step and guarantees the stability of the original large-scale system. Finally, the proposed method is applied to a physical setup-a soft robotics platform-to show its feasibility.