Multi-Sensor Fusion Based Mobile Robot Localization In Indoor Environment

Localization is a foundational issue for mobile robots,especially in autonomous navigation tasks.Due to the great instability in some scenarios,most single-sensor based methods are not able to operate stably in long-term.And how to utilize multi-sensor fusion to obtain a more stable and accurate performance is also a tough issue in the research.This thesis researches multi-sensor fusion based localization methods which focus on the long-term and stable application for indoor mobile robots.In the end,an online calibration algorithm and a general multi-sensor fusion localization framework are proposed,a localization system with the different combination of sensors in diversity scenarios is implemented also.The main contributions are as follows:1.Two single-sensor based localization methods are developed respectively for stereo camera and 2D LiDAR.It is pointed out that the single sensor based method has great instability in some scenarios,which indicates the necessity of multi-sensor fusion localization.2.A time-synchronized tightly coupled optimization based multi-sensor online calibration method is proposed.The incompletely rigid changes in the environment and the robot assembly will affect the extrinsic calibration of sensors and the localization accuracy.This thesis proposes an online calibration and localization method which is aimed at the issues above.First of all,the expression of robot pose is unified.After that,we propose a general sensor pre-integration theory for fusing sensors at different sample rates to adapt the iterative adjustment process in optimization.Finally,the online calibration is carried out based on the off-line calibration.Being added into the fusion framework as an optimization variable,the calibration parameter is modified in real time while localizing,which flourishes high mobility and environmental adaptability.The experimental results validate that the online calibration algorithm is robust and valuable especially in the fusion based localization with camera and LiDAR,which is able to reduce the error from initial value and runtime.3.A general localization framework with graph optimization is proposed.Some multi-sensor fusion localization methods against three different combinations of sensors for indoor mobile robot are implemented based on the proposed general fusion framework.The experimental results show that the multi-sensor fusion method is more accurate and robust than the single-sensor based method,and the odometer,gyroscope,stereo vision fusion localization method is proved to be the best performance on the mobile robot.