Research On Mapping And Localization Of Indoor AGV Based On LADAR

As a kind of automatic logistics equipment, AGV(Automated guided vehicle) is increasingly being used in material handing and assembly process. Navigation technology is one of the core technologies of AGV. The Navigation technology based on LADAR which can achieve high positioning accuracy, high routing flexibility and high level of intelligence is consistent with the development trend of AGV navigation technology. Although laser navigation has been successfully applied in AGV field, in existing navigation technology based on LADAR, there are problems of dependence on reflectors, low precision in mapping and inconvenience in AGV applying caused by that initial localization requires human intervention. Based on the review of the key technologies and the development of laser navigation, the paper carried out basic research of laser navigation technology in the background of structured indoor environment application around three key works: environmental perception, mapping and localization.On the basis of analysis and comparison of the respective characteristics of segment feature and point feature, corner points and break points are choosen to present the envirement map. In view of the probem that thus feature can not be accurately and stably extracted directly from LADAR scan datas, a strategy is adopt that segment feature is extracted and then on this basis point feature is extracted. On extracting feature segments, a method based on split-merge is presented since IEPF(Iterative End Point Fit) easily results in over-split and under-split, and then on the basis that the uncertainty model of LADAR scanning is established and its estimating method is given, a method on estimating parameters and uncertainty of feature segments is derived. On extracting feature points, in order to increase the number of feature points, on the basis of the points in some cases which Connette considers, the points in two new cases are increased and a recgonition method is given, and then according to the form of feature points methods of estimating the position and its uncertainty of corner points and break points are derived respectively.Considering normal EKF(Extended Kalman Filter) SLAM’s(Simultaneous Localization and Mapping) problem of inconsistency, an EKF SLAM method fusing absolute orientation information is presented. The method divided SLAM into two layer(inner and outer) structions of EKF filtering. The inner layer EKF predicted AGV’s pose with encoder information and corrrrected AGV’s pose with compass information. The outer layer EKF predict AGV’s pose with the output of the inner layer EKF and then corrected AGV’s pose and map with LADAR datas. An adaptive method estimating system noise based on IAE(Innovation-based adaptive estimation)-CMAC(Cerebellar Model Articulation Controller) is presented, since it is difficult to obtain statistical characteristics of EKF SLAM system. The method approximates to a mapping function from velocity and angular velocity to the statistical characteristics of system noise with CMAC, and gains the learning samples of CMAC with IAE, and adaptively estimates the system of EKF SLAM when training CMAC.Considering that initial pose needed when poseture tracking with EKF can not be given easily, a Markov localization method that can globally localize mobile robots in feature map is presented. The method includes time update and measurement update. In the stage of time update, in order to impement translation operation of nonintegral multiple of a unit grid a method based on operating in the frequency domain is presented. After that the discretized three dimensional pose space is reduced into the two dimensional one, the method transforms the confidence image into frequency domain from spatial domain, and then according to the relationship between the images in the spatial and frequency domain finishs translation operation of nonintegral multiple of an unit grid and blurring operation, and finally transforms the confidence image backto spatial domain from frequency domain. In the stage of measurement update, in order to calculate likelihood without data association and efficiently take advantage of recursive Bayesian estimation, a method of measurement update based on Gaussian kernel smoothing is presented. The method transformed the landmarks extracted into continuous form through Gaussian kernel fuction, and then discretized the virtual continuous environment contour with equal angle intervals, and at the last calculate a likelihood by establishing the measurement likelihood model.In order to verify the theories and methods mentioned above, on the basis both of the EKF SLAM method fusing absolute orientation information and mobile robotic global localization method based on Markov are verified through the simulation, an experimental platform of laser navigation AGV is designed and developed. Sensor datas from a LADAR, an encoder and a compass are gathered in indoor environment by the platform and experiments of feature extraxting, mapping and localization are finished. Experiment results show that the theories and methods presented in this paper are valid.Finally, the main innovation and research achievements of the thesis are summarized, and the futher work of the researches on the subject is expected.