Random weighting estimation for systematic error of observation model in dynamic vehicle navigation

The Kalman filter requires kinematic and observation models not contain any systematic error. Otherwise, the resultant navigation solution will be biased or even divergent. In order to overcome this limitation, this paper presents a new random weighting method to estimate the systematic error of observation model in dynamic vehicle navigation. This method randomly weights the covariance matrices of observation residual vector, predicted residual vector and estimated state vector to control their magnitudes, thus governing the random weighting estimation for the covariance matrix of observation vector. Random weighting theories are established for estimations of the observation model's systematic error and the covariance matrices of observation residual vector, predicted residual vector, observation vector and estimated state vector. Experiments and comparison analysis with the existing methods demonstrate that the proposed random weighting method can effectively resist the disturbance of the observation model's systematic error on the state parameter estimation, leading to the improved accuracy for dynamic vehicle navigation.