New developments in nonlinear systems estimation

The Extended Kalman Filter (EKF) has unquestionably been the most widely used estimation algorithm for nonlinear systems. However, the EKF is based on first-order Taylor approximations of state transition and observation equations about the estimated state trajectory. The EKF provides an insufficiently accurate representation in many cases, and it is difficult to implement and tune. Many of these difficulties arise from its use of Taylor linearization. To overcome this limitation, the interpolation filtering, unscented filtering, particle filtering and neural network filtering are developed as new nonlinear estimation methods in these years. From the two developmental ways (non-Taylor linearization linear transformation and nonlinear transformation) of nonlinear estimation innovation, these new approaches' motivation, operational principle and performance is reviewed in details, furthermore, some underlying assumptions, flaws and future challenges are pointed out. Although these new algorithms represent sufficient superiority to be applied in many highly nonlinear filtering and control applications, it is important to recognize that the selection of these algorithms should suit measures to local conditions.