Improving Adaptive Kalman Estimation in GPS/INS Integration

Abstract

The central task of GPS/INS integration is to effectively blend GPS and INS data together to generate an optimal solution. The present data fusion algorithms, which are mostly based on Kalman filtering (KF), have several limitations. One of those limitations is the stringent requirement on precise a priori knowledge of the system models and noise properties. Uncertainty in the covariance parameters of the process noise (Q) and the observation errors (R) may significantly degrade the filtering performance. The conventional way of determining Q and R relies on intensive analysis of empirical data. However, the noise levels may change in different applications. Over the past few decades adaptive KF algorithms have been intensively investigated with a view to reducing the influence of the Q and R definition errors. The covariance matching method has been shown to be one of the most promising techniques. This paper first investigates the utilization of an online stochastic modelling algorithm with regards to its parameter estimation stability, convergence, optimal window size, and the interaction between Q and R estimations. Then a new adaptive process noise scaling algorithm is proposed. Without artificial or empirical parameters being used, the proposed adaptive mechanism has demonstrated the capability of autonomously tuning the process noise covariance to the optimal magnitude, and hence improving the overall filtering performance.