Robust Multi-layer Vehicle Model-Aided INS Based on Soft and Hard Constraints

The navigation accuracy is an important performance indicator for intelligent vehicles. Errors of the inertial navigation system (INS) diverge fast in the electromagnetic signal denial environment, and the vehicle model aid is a common approach to solve it. However, the accuracy of the vehicle model will be influenced by the road condition, vehicle motion state, and other interferences, so the precision degradation and gross error are frequent. Against this problem, this paper proposes the robust multi-layer vehicle kinematics model-aided INS. First, the gross error detection and isolation method is proposed based on the singular value decomposition and chi-square test, and problems caused by the inadequate measurement redundancy are analyzed and solved by the "model redundancy". Second, the multi-layer vehicle model-aided inertial navigation system is proposed, which contains the sensor layer, the system layer I, and the system layer II. In the sensor layer, the soft and hard constraints of the vehicle model are defined, and the vehicle velocities are estimated with the constraints. Then, in the system layer I, the vehicle velocities from the sensor layer are used to estimate and correct INS errors. At last, the position is updated in the system layer II with the corrected attitude and velocity, so that the navigation performance can be improved. The simulation and field experiments proved the proposed method has the higher navigation accuracy and the better robustness against the gross error and high-dynamics situation, that its position error is less than 20 m in the seven-minute field test.