Multi-Kernel Maximum Correntropy Kalman Filter for Orientation Estimation

Inertial measurement units (IMUs), composed of gyroscopes, accelerometers, and magnetometers, have been widely used in the fields of human motion animation, rehabilitation, robotics, and aerospace. However, their performances degenerate remarkably with external acceleration and magnetic disturbance. To handle this issue, we employ a multi-kernel maximum correntropy Kalman filter (MKMCKF) to suppress the adversarial acceleration and magnetic disturbance and use Bayesian optimization (BO) to explore the optimal kernel bandwidths. We validate our algorithm in a set of experiments with different levels of disturbance. Results show that the proposed method is significantly better than the traditional error state Kalman filter (ESKF) and the gradient descent (GD) method, and its root mean square error (RMSE) is less than $0.4629^{\circ }$ on the roll and pitch even under the worst testing case.