Research On Nonlinear Bayesian Filtering Based SLAM Algorithms

With the rapid development of information technology,mobile robots have been widely used in many fields,such as home service,medical health,industrial production,underwater exploration,aerospace,and military.In the application of mobile robot,autonomous navigation is the key problem to be solved first.In the complex unknown environment,it is difficult for mobile robot to obtain accurate position information.Simultaneous localization and mapping(SLAM)technology is an effective method to solve the robot position estimation in unknown environment.In this method,robot sensors are applied to perceive the surrounding environment,Bayesian statistical theory is used to estimate the robot pose and feature landmark position,and the environment map is gradually constructed and perfected,then the autonomous positioning and navigation of mobile robot are realized.However,there are sometimes large external interference noise,model uncertainty,sensor error and observation outliers in practical application,which leads to the degradation of state estimation accuracy and robustness of SLAM system,and seriously affects the robot autonomous navigation.To solve the above problems,the nonlinear Bayesian filtering based SLAM algorithms are deeply studied in this dissertation,and several robot state estimation methods are proposed to improve the performance of robot localization in SLAM.The main innovative works of this dissertation are as follows:(1)Aiming at the problems of low robustness and large computational load for the unscented Kalman filter based SLAM algorithm,an improved Schmidt orthogonal unscented Kalman filter based SLAM algorithm is proposed.The sampling strategy based on Schmidt orthogonal transform is first introduced to reduce the sampling points of unscented transform.Then,according to the idea of strong tracking algorithm,a fading factor is obtained by using the orthogonal principle of residual vector to modify the prediction covariance matrix,which can improve the tracking ability of system.Besides,the square root filter is used to avoid the divergence problem caused by non-positive definite covariance matrix.The proposed algorithm reduces the computational amount,and improves the robustness of system.In addition,to solve the performance degradation when the noise covariance is unknown,a Masreliez-Martin unscented Kalman filter based SLAM algorithm is proposed,where Masreliez-Martin filter is combined with unscented Kalman filter(UKF),the Huber function and moving window are applied to improve the system robustness and positioning accuracy,and the adaptive adjustment scheme based on observation residuals is applied to modify the prediction noise covariance matrix through adaptive factors,so the algorithm can also have good SLAM positioning performance when the noise covariance is unknown.(2)Aiming at these problems of poor robustness and low positioning accuracy in UFast SLAM,an improved adaptive unscented Fast SLAM algorithm is proposed.Firstly,the unscented Kalman filter is improved,where the time-varying noise statistical estimation is applied to deal with the unknown problem of system noise,the QR decomposition of matrix is utilized to avoid the non-positive definite problem of covariance matrix,the SVD decomposition is utilized to deal with the numerical stability caused by the inversion of covariance matrix in gain calculation,the Huber cost function is employed to suppress the observation outliers,and the covariance matrix is adjusted by fading factor to improve the tracking ability of the system.Then,the improved adaptive unscented Kalman filter is used to obtain the mean and covariance of the proposal distribution function,and realizes the importance sampling for robot pose state;the improved genetic algorithm is adopted for resampling to suppress the sample impoverishment problem by increasing the particle diversity.The proposed SLAM algorithm improves the positioning accuracy,suppresses the influence of state sudden changes and measurement outliers,and improves the robustness of SLAM algorithm.(3)To further enhance the robustness against interference noise for unscented Fast SLAM(UFast SLAM),an improved H-Infinity unscented Fast SLAM algorithm is proposed.Aiming at the system state estimation and noise suppression,the corresponding cost function is constructed using the idea of H-Infinity filter.Based on the minimization criterion of maximum error,an improved H-Infinity unscented Kalman filter algorithm is provided,where the time-varying noise estimator is utilized to deal with the unknown problem of system noise,and the Huber function is utilized to suppress the observation outliers and improve the robustness of system.Then,the improved H-Infinity unscented Kalman filter is used to obtain the mean and covariance of the proposal distribution function and realizes the importance sampling for robot pose state,and the adaptive genetic algorithm is adopted as resampling to improve the particle diversity.The proposed SLAM algorithm can improve the location performance of SLAM system and has good robustness.(4)Aiming at the poor performance of SLAM algorithm in the non-Gaussian noise environment,an improved UFast SLAM algorithm with generalized correntropy loss is proposed.Considering the non-Gaussian noise characteristics of impulse noise and observation outliers,the generalized Gaussian distribution is used as the kernel function,which can construct the cost function based on generalized correntropy,and the generalized correntropy loss function and fixed point iteration technique are employed to improve the unscented Kalman filter.Then,the unscented Kalman filter with generalized correntropy loss is used to obtain the mean and covariance of the proposal distribution function,and realizes the importance sampling for robot pose state.Based on the adaptive mutation probability,the improved genetic algorithm is adopted as the resampling strategy to increase the effectiveness and diversity of particles.This method can effectively suppress the influence of non-Gaussian noises,improve the robustness of SLAM algorithm.