Composite Non-Gaussian Filtering With Intermittent Inputs: Algorithm Design and Applications

Abstract

We investigate the state estimation problem for a general class of nonlinear systems subject to multisource heterogeneous disturbances including dynamic uncertainties, stochastic noises, and intermittent inputs. In particular, the intermittent input term in the state transition equation, stemming from packet loss of the control signal without acknowledgment or occasional failure of the actuator, is an important source of uncertainties in practice. Nevertheless, this intermittent input term has received relatively little research attention. In this article, we use a Bernoulli binary indicator to describe the intermittent availability of input signals. On this basis, a composite non-Gaussian filtering scheme is developed, which consists of a variational Bayes-adapted disturbance observer (DO) to estimate the joint distribution of dynamic uncertainty and binary indicator, and a robust particle filter (PF) with a modified proposal distribution to track the non-Gaussian posterior probability density function (pdf) of the state variable. Compared with the existing methods, our approach can make more effective use of the heterogeneous characteristics of dynamic uncertainties, stochastic noises, and intermittent inputs, thus producing refined antidisturbance estimation results. Furthermore, the proposed scheme is applied to attitude estimation of quadrotor unmanned aerial vehicles (UAVs) subject to packet loss of the controller-to-actuator channel, enhancing the resilience of the UAV attitude determination system against unreliable control signal transmission.