In multirotor unmanned aerial vehicle (UAV) GNSS/INS integrated navigation systems, a single filter such as the extended Kalman filter (EKF) or the error-state extended Kalman filter (ESKF) is commonly adopted. However, both methods have inherent performance limitations. The EKF suffers from significant linearization errors in highly nonlinear flight scenarios, leading to degraded estimation accuracy. Although ESKF achieves higher precision during steady flight, its model assumptions may no longer strictly hold during aggressive maneuvers, causing performance degradation in complex flight missions. To address the limitations of using a single filter, this study proposes a dynamic filter selection strategy under the interaction multi-filter (IMF) framework. The approach builds on the interactive multiple model (IMM) method and establishes a cooperative mechanism between EKF and ESKF. By computing the filter likelihoods at each time step and updating the probability switching matrix, the framework adaptively selects the optimal filter based on the current flight conditions. Simulation results demonstrate that the proposed IMF-based strategy effectively avoids the performance bottlenecks of individual filters. In highly nonlinear environments, it reduces linearization errors and suppresses divergence trends; compared with traditional ESKF, the proposed algorithm 3D RMSE is reduced by 57.2%, compared with the adaptive robust EKF (AREKF), the proposed approach reduces positioning errors by up to 21.3%. The results confirm that IMF-based adaptive switching between EKF and ESKF yields a robust, high-precision solution for UAV navigation in complex operational scenarios.
Zhang et al. (Thu,) studied this question.