| Study of service robots in indoor environment has become a hot topic since the late 80's in the last century, and has been made great achievements in the theory and practice. As the basis of navigation, task planning, mobile operation and localization have drawn greater attention. Purpose of this thesis is to enhance the reliability and portability of robot system, and to improve the accuracy and reliability of indoor localization. Then begin with the establishment of service robot systems, based on it indoor mobile localization is done, those as follows.A new type of service robot system framework and modeling are given based on modular technology,whose ideas covered the five subsystem. The reliability problems of systems and related solutions are studied, and finally related modulars are designed to build the system. At the same time, the robot motion and sensors are modeled, tested or emulated, considering the robot motion model, odometer model, the infrared tag sensor and laser sensor observation model.The SLAM robot localization algorithm description is given based on Bayesian theory, the factorization of the full state Fast-SLAM2.0 being derived. Furthermore, based on analysis of limitation and key factors, a grid-based FastSLAM2.0 algorithm is proposed. According to the elmulation of the algorithm under the conditions of dense and sparse features environments, the FastSLAM2.0 algorithm is summarized and the related basic theory is studied for the new algorithm. Besides, grid-based FastSLAM2.0 is emulated.Based on the square root unscented Kalman filter, the new simultaneous localization and mapping) algorithm, that is proposed SRUKFSLAM in which square root matrix is propagated in order to ensure that the status of non-negative definite matrix. The square root unscented Kalman filter is used as estimator of the state of the robot and simultaneously estimates infrared tags as landmarks, which achieve higher state estimation accurancy than FastSLAM2.0. Besides,Not only the new algorithm is derived and analyzed,but also the relevant compared emulator and experiments are done. The performances of compared experiments such as the robot state estimating, landmarks estimating and deviation of close loop verified that the stability and the accuracy of SRUKFSLAM are better than FastSLAM2.0.Based on MbICP, considering dynamic features, MbICP-based strategy is proposed which can detect moving targets and MbICP denosing for improving the robot incremental state estimation.Finally, in a dynamic environment, the two assumptions in view of features are improved. Furthermore, dynamic environment objectives are factored according to those assumptions, first-order Markov assumptions and Bayesian theory. As a result, layered localization comprehensive algorithm for dynamic environment is proposed, which fuses infared tags, laser sensor, and MbICP strategy. The algorithm summarized as three layers. The first layer executing moving target tracking; the second used as SLAM1 local metric map and the third layer performing SLAM feature map. Note that the third layer-feature map is still built by SRUKFSLAM, when no landmarks observed, the robot state updated by calling MbICP strategy. The compared experimental results verify the effectiveness and stability of the comprehensive layered localization algorithm. In this thesis, a mobile robot platform is developed with modular idea which improves the stability and portability by study on the systems and is beneficial to robot industrialization, while the robot motion model and sensor observation are studied to provide a favorable reference for indoor localizaion. Considered the static and dynamic environmental factors, different algorithms are proposed in this thesis. Emulator experiments of those algorithms verify that the algorithms can improve the accuracy and stability of indoor robot localization and possess a practical reference value.
|
PDF Full Text Request