Research On The Algorithm Of Simultaneous Localization And Mapping For Robot In Complex Environment

Mobile robots have played more and more important roles in military,industry,agriculture and people's daily life.Simultaneous Localization And Mapping(SLAM)technology is the basis of autonomous for robots,and also is the key to ensure robots completing specified tasks in various environments.However,with the increasingly complex application environment,mobile robots still have many shortcomings in positioning accuracy,algorithm robustness and adaptability to complex scenes,which limits the application of robots in many occasions.In order to complete the pose estimation and mapping accurately,robustly and in real-time for robot in complex environments,this thesis studies all aspects of SLAM technology.The main contributions are as follows:(1)In order to improve the accuracy of pose estimation fusing visual and inertial information for robot,a novel visual-inertial odometry(VIO)algorithm is proposed,which is based on matrix Lie group and Cubature Kalman filter framework.The proposed algorithm is constructed by combing invariant theory and Cubature Kalman Filter.The state variable is constructed by a high-dimensional matrix Lie group,which exactily represents the uncertainty of rotation and IMU bias.In order to propagate the uncertainty accurately and reduce the linearization error of the VIO system,a new method of cubature transformation running in manifold space is proposed.The VIO is completed using invariant Cubature Kalman Filter,which effectively improves the accuracy of pose estimation for mobile robot.(2)In order to improve the accuracy and robustness of estimation for VIO in texture-less and illumination-changing scenes,a VIO algorithm combining line features and invariant Cubature Kalman filtering is proposed.In the proposed algorithm,line features are added into the state variable constructed by high dimensional Lie group using the method of line endpoints,which realizes the fusion of line features for VIO.The algorithm studies the cubature transformation on manifold space to transform the line uncertainty,and deduces the update process using observation function built by the back projection of tracking lines.A robust pose estimation is realized under the invariant Cubature Kalman Filter framework.A selecting method of matching lines with geometric constraint is used for improving the accuracy and reliability of matching lines.In the filtering process,a new scheme of line management is provided,which effectively solves the problem of partial observation in line measurement.(3)A illumination robust loop-closure detection algorithm with pose constraints is proposed to solve the problem of visual confusion and real-time problem in visual loop-closure detection.The proposed algorithm adopts odometry information to constraint the detection process,and a distance function including pose uncertainty is built.The selection strategy of loop-closure candidate region is proposed,which solves the problem of candidate area loss caused by odometer drift and effectively reduce the calculation of loop-closure detection.In the candidate region,the DIRD is used to complete the loop-closure detection,which improves the illumination robustness.(4)An improved probability hypothesis density filter is proposed,which is applied in situations of high clutter and data association ambiguity.In this algorithm,the cubature rule is utilized to calculate Gaussian weighted integral of the nonlinear function,and the performance of pose estimation is improved by improving the accuracy of particle weight calculation.In the process of GM-PHD update,square-root cubature Kalman filter(SCKF)is utilized in the calculation of measurement likelihood and Gaussian component's weight,which guarantees the symmetry and positive semi-definiteness of the covariance matrix and improves numerical stability and accuracy.Figure: 76,Table: 13,Reference: 133.