Research On Mobile Robot Localization Based On Multi-sensor Information

Mobile robot localization is the basic and most important function of autonomous navigation, and also a challenging main research subject which gets most of the attention in the field of the robotics. The dissertation centers around pose tracking in known structured environment, Multi-sensor information fusion algorithm is applied to the integrated positioning system based on odometer and multi-sonar. The paper includes contents as follows:Firstly, considering structural features and functional requirements of the self-designed mobile robot, an idea of double CPU based on DSP TMS320F2808 and ATmega32 is adopted. The dissertation gives detail introduction on hardware design of the system.Secondly, work principle of odometer and sonar is introduced and error resource is analyzed, on the basis of which the arc kinematic model of odometer and the measurement model of sonar for line feature are made.Thirdly, to address the accumulative error problem caused by odometer, a localization algorithm, which employs extended kalman filter (EKF) to fuse odometer readings and sonar data, is proposed. It first employs odometer and given map to obtain predicted value of the features, then extracts the line features in environment from sonar data and periodically corrects the cumulative error of odometer via matching and updating process between real observation and prediction to improve position precision.Fourthly, through the analysis of the effects of observation noise's statistic characteristics on the performance of EKF, a fuzzy logic based adaption tuning algorithm is presented. This method can on-line adjust the observation noise covariance by monitoring the coincidence of the actual covariance with the theoretical covariance of residuals, so as to overcome the bad effects of incomplete knowledge about it and on-line improve the performance of localization method.