Linear unknown input-state observer for nonlinear dynamic models

This paper proposes an unknown input observer for nonlinear systems with input decoupling via system invertibility. Starting from a suitable reformulation of the model of a generic nonlinear system, obtained by merging all system uncertainties with respect to an appropriate nominal linear model into a disturbance vector, the proposed observer can asymptotically copy both the system state and unknown inputs, even in the presence of measurement noise. Formal proof of the estimate convergence is demonstrated analytically. A comparison of the proposed method with existing solutions is shown in simulation, and the method’s effectiveness in real-world scenarios is demonstrated by experimental results on a soft articulated robot.