Decoupled Multikernel Correntropy-Based Invariant Extended Kalman Filter for IMUs

Inertial measurement units (IMUs) are widely utilized in state estimation applications. However, their performance can deteriorate significantly in the presence of unknown disturbances. This article proposes a decoupled multikernel correntropy-based invariant extended Kalman filter (DMKCIEKF) for robust IMU orientation estimation. By integrating a multikernel correntropy-based filtering mechanism with Lie group theory, the proposed method effectively suppresses disturbances in accelerometer and magnetometer measurements. Moreover, the magnetic vector is projected onto the horizontal plane such that magnetometer measurements contribute exclusively to yaw estimation, while their influence on roll and pitch is eliminated. Combined with a time-separated measurement update scheme, the proposed approach accommodates the rate mismatch between accelerometer and magnetometer measurements. Experiments were conducted to evaluate the orientation estimation performance of the DMKCIEKF under both static and dynamic conditions, with and without magnetic interference. The experimental results demonstrate that the proposed DMKCIEKF consistently outperforms existing estimators in both static and dynamic scenarios. In particular, the root-mean-square-error (RMSE) is kept below 0.709^◦ for roll and pitch angles, even under severe magnetic interference.