Research On Localization Technology For Mobile Robots

As an important research area of robotics, localization is the critical step towards autonomous navigation of mobile robots, and promises to improve automatization level. Position-estimation methods can be categorized into two groups: absolute and relative positioning. Absolute positioning requires that robot determines its position when initial position isn't given. Relative positioning is the process of position determination on condition that robot is told its initial position, which is the main research topic of localization. In this dissertation, technologies related to relative positioning are focused on, including calibration of systematic parameters of robot, identification and correction of wheel-slippage, modeling of odometric noises and adaptive estimation of noise statistic character, matches between laser scanning data and map and posture error compensation of robot based on laser scanning data. Correct systematic parameters of robot are the prerequisite to accurate localization. UMBmark Calibration is the most widely used method to calibrate systematic parameters of robot. However, it can't effectively correct the orientation error due to the difference between the average of two actual wheel diameters and the nominal wheel diameter when experimental robot has poor control precision, which deteriorates calibration result. This dissertation analyzes UMBmark Calibration and presents an improved method to measure systematic errors of low-precision mobile robots. The improved method considers unequal wheel diameters, uncertainty about the wheelbase and the difference between the average actual wheel diameter and the nominal wheel diameter as the main error sources of odometry and defines a new model for odometry. Detailed algorithm is proposed to calculate the systematic parameters of a mobile robot and the corresponding correction factors. Experimental results show that the improved method has better calibration effect than UMBmark Calibration. Correcting errors from wheel-slippage is the key technology of precise localization. Under the constraint that electronic compass isn't distorted by magnetic interference, this dissertation is based on the information from encoders and electronic compass, and proposes a novel algorithm to identify and calibrate wheel-slippage when robot performs straight-line motion. Based on experimental data, statistic theory is implemented to define a model for wheel-slippage. This model can judge whether driving wheels have slippages when robot navigates. Based on the model, an algorithm is presented to determine which driving wheel has slippage and calibrate the resulting orientation error and displacement error when wheel-slippage occurs, and an Indirect Kalman Filter (IFK) is further introduced to calculate the actual diameters of driving wheels and wheelbase of robot. Experimental results show that localization accuracy can be effectively improved by correcting wheel-slippage, and the diameters of driving wheels can be precisely determined by IFK. Adaptive estimation of the statistic character of odometric noises is a difficult problem for localization. In this dissertation, noise sources of odometry are analyzed, the errors due to encoder resolution are taken into account and a model for odometric noises is defined. Based on the model, equations are deduced to estimate the statistic character of odometric noises real-time and adaptively when mobile robot moves. Recently, laser-based localization has been a topic of interest in position-estimation. It is an important step for laser-based localization to extract line segment characteristics by rapidly matching laser scanning data with map. By defining g-weighted Hough transform, Effective Area of Plane (EAP) is put forward in this dissertation. Line segment characteristics are extracted from laser scanning data, and simultaneously the matches between line segments and map are realized, which greatly improves process efficiency of laser scanning data. Simulation results show that line segment information can be effectively and rapidly extracted by our method. Based on the line segments extracted from laser scanning data, weighted least squares are utilized to calculate orientation error, and point-to-line least squares and point-to-point least squares are proposed to compute position error. By comparison, it is found that point-to-point least squares have better calibration effect than point-to-line least squares. Extended Kalman Filter (EKF) is implemented to fuse the information from encoders and laser range finder to locate mobile robot. Experimental results show that orientation error and position error can be effectively corrected and robot's pose is also accurately tracked by EKF.