EKF for simultaneous vehicle motion estimation and IMU bias calibration with observability-based adaptation

Autonomous driving systems require precise knowledge of the vehicle motion states such as velocity and attitude angles. In this work, an extended Kalman filter-based approach is proposed for a simultaneous vehicle motion estimation and IMU bias calibration. The estimation scheme relies on the combination of a kinematic model-based estimator with dynamic model-based measurement equations. As observability issues are closely linked to the estimator's accuracy and reliability, an exhaustive observability analysis is vital. It is shown that the analysed system is nonuniformly observable. Nevertheless, universal inputs that secure the states' distinguishability are analytically found. In order to prevent the estimates from diverging when distinguishability is not guaranteed, an observability-based adaptation is proposed. The effectiveness of the method is illustrated on real data collected from a regular city drive, where a significant enhancement of the motion state estimates' accuracy is achieved.