An Adaptive Kinematics Model-aided Navigation Method For Unmanned Ground Vehicles

The vehicle kinematics model aided inertial navigation system (VKM-AINS) is an effective solution for the autonomous navigation in the unmanned ground vehicle (UGV). However, traditional VKM-AINS methods suffer from fast error divergence and unstable VKM accuracy, leading to degraded performance in situations of long-duration GNSS failure, challenging road conditions, and high-dynamics motion. To address these challenges, this paper proposes a novel VKM-AINS incorporating the state transformation (ST) error model and variational Bayesian (VB)-based adaptive estimation algorithm. First, the error coupling problem is analyzed, which is one of the root causes for the fast error divergence, and the state transition function and observation function are developed with the ST error model to decouple the effects of the velocity and attitude error. Second, the issue of the time-varying observation noise covariance matrix, arising from the unstable VKM performance across different motion states, is analyzed. A VB-based filter is proposed to simultaneously approximate the distributions of observation noise and state variables, enabling optimal estimation despite the varying driving states. Field experiments validate the performance improvement of the proposed method, demonstrating a 19.07% reduction in horizontal position error compared to conventional VKM-AINS, while maintaining real-time computational efficiency.