In this paper, we propose a new method for modeling and output regulation of nonlinear systems with disturbance noises. We can get a semi-global error system between a real plant and a simplified model as a LPV system with two-step system identification method including L0-quasi norm minimizing sparse algorithm. LPV error system can produce compensating input for conventional homogeneous output feedback law, which enables us to release the assumption which isn't satisfied by the real plant, and globally regulate the real plant without its detail information. Furthermore, we propose a new nonlinear filtering method for the output feedback. We conventionally use Unscented Kalman Filter (UKF) which can't reduce non-Gaussian noise effect to nonlinear systems effectively. Instead UKF, we propose Robust Unscented Kalman Filter (RUKF). This filter consists of UKF and Robust Kalman Filtering algorithm using L1 optimization. RUKF can handle non-Gaussian, especially outlier of measurement value, which causes homogeneous output feedback to diverge. We confirm the effectiveness of this method with numerical simulations.