Research On Filtering Method For Vision-Aided Navigation System In Asteroid Exploration

Asteroid exploration becomes an important development direction in the field of deep space exploration in the 21 st century because it is significant for revealing the origin of the Solar System and monitoring potential threats to the Earth.Precise autonomous navigation is an indispensable technology for future asteroid prospectors to successfully carry out missions during approach phases,such as orbiting and orbit transfer.The vision-aided navigation(VAN)based on the inertial measurement unit and navigation camera has the advantages of high accuracy and strong autonomy and has extensive application prospects in asteroid exploration.For the vision-aided navigation system(VANS)with strong nonlinearity,its navigation accuracy largely depends on the nonlinear filtering algorithm used to fuse the measured data of the visual and inertial navigation.This paper takes asteroid orbiting exploration as background and conducts in-depth research on filtering method for VANS.The main contents are listed as follows:First,the measurement model of the inertial navigation system(INS)and visual navigation system(VNS)and landmark point selection method based on position dilution of precision(PDOP)for VNS are provided,and the state transition equation and measurement equation of the VANS are established.Then,by using the visual measurement,the extended Kalman filtering is utilized to correct the navigation error of INS to obtain the accurate position,velocity and attitude of the prospector with respect to the asteroid.Finally,mathematical simulations are conducted to validate the validity and effectiveness of the VAN model and landmark point selection method for visual navigation.To improve the estimated accuracy and convergence rate of the navigation filter,the nonlinear Gaussian filtering based on the deterministic sampling is introduced to deal with the problem of state estimation from nonlinear systems.The high-degree cubature Kalman filtering based on the fifth-degree spherical-radial cubature rule is derived by using the Genz's integration method and moment matching.By employing the Taylor series expansion of a multivariable function and the computational complexity,the performance of the HCKF,cubature Kalman filtering(CKF)and Gauss-Hermite quadrature filtering(GHQF)are synthetically evaluated in terms of estimated accuracy and calculated quantity.The theoretical analysis demonstrates that the estimated accuracy of HCKF is equivalent to GHQF and superior to CKF,while the calculated quantity of HCKF is far less than that of GHQF.The vision-aided navigation filter based on the HCKF is designed.Through simulation,the HCKF is a nonlinear Gaussian filtering algorithm with advantages of estimated accuracy and computational efficiency.Under the influence of different lighting conditions,the measurement noise of navigation camera often presents as contaminated Gaussian distribution with unknown covariance.As for the unknown covariance of the measurement noise,Gaussian distribution and Inverse Wishart distribution are used to respectively model the system state and measurement noise covariance in the conjugate-exponential field,and the system state and noise covariance is recursively estimated by using variational Bayesian theory.In order to suppress the influence of non-Gaussian noise on the estimation of the system state and measurement noise covariance,the generalized maximum likelihood estimation technique proposed by Huber is researched and utilized to modify the update step of the nonlinear Gaussian filtering,and the variational Bayesian adaptive high-degree cubature Huber-based filtering(VBAHCHF)is derived.The vision-aided navigation filter based on the VBAHCHF is designed,and the mathematical simulation is conducted to verify the advantages of the VBAHCHF algorithm in terms of adaptability,robustness and estimation accuracy under the case of contaminated Gaussian noise with unknown covariance.On the other hand,due to serial interface conversion between the navigation system and onboard computer,the visual measurement data transmitted to the data fusion center may be randomly delayed by multiple sampling times.To address the estimation in such a situation,the multiple-step randomly delayed nonlinear system is firstly established by using multiple Bernoulli random variables to construct the relationship between the actually received measurement and the ideal measurements.Then,the likelihood function of the filter is computed by marginalizing out the delay variables to extract accurate information from the delayed measurement,and the multiple-step randomly delayed high-degree cubature Kalman filtering(MRD-HCKF)is derived.The multiple-step randomly delayed variational Bayesian adaptive high-degree cubature Huber-based filtering(MRD-VBAHCHF)is proposed by embedding the online estimation of measurement noise covariance based on variational Bayesian theory and the robust estimation on the basis of Huber technique into the MRD-HCKF.Finally,the vision-aided navigation filter based on the MRD-VBAHCHF is designed,and the mathematical simulations are carried out to demonstrate that the MRD-VBAHCHF outperforms the HCKF,MRD-HCKF and VBAHCHF for the nonlinear system whose measurements are randomly delayed by multiple sampling times and corrupted by contaminated Gaussian noise with unknown covariance.