Variational Bayesian adaptive high-degree cubature Huber-based filter for vision-aided inertial navigation on asteroid missions

Vision-aided inertial navigation (VAIN) is a prospective technique for determining the pose of the spacecraft during asteroid missions. The VAIN system can fuse the inertial and visual data by employing the high-degree cubature Kalman filter (HCKF) because it can accurately handle non-linear problems. However, the visual measurements can be corrupted by non-Gaussian noise with unknown time-varying covariance, resulting in severe degradation of the HCKF. To improve the navigational accuracy of the spacecraft in these situations, the authors propose a novel adaptive robust HCKF known as variational Bayesian (VB) adaptive high-degree cubature Huber-based filter (VB-AHCHF). In the novel algorithm, the fifth-degree cubature rule and VB theory are combined to estimate the state and track the non-stationary statistical characteristics of the measurement noise. In addition, utilising the M-estimation, which is defined as the Huber technique, it modifies the update step of the formal Bayesian filtering. Therefore, the VB-AHCHF can exhibit adaptability and robustness to the covariance uncertainty and non-Gaussianity of the measurement noise. Their simulation results show that the estimation accuracy of VB-AHCHF, as well as its adaptability and robustness, is superior to all state-of-the-art algorithms, e.g. HCKF, high-degree cubature Huber-based filter, and the VB adaptive HCKF.