Research On Uncertainty Treatment Of Mobile Robot Localization

This dissertation is supported by the key project of the National NaturalScience Foundation of China under grant no.60234030, Research on Theoriesand Methods of Navigation Control for Mobile Robots under UnknownEnvironments. As one part of the project, the dissertation is developed with thelocalization problem in mobile robot navigation. Combined with "MobileRobot 1 of Central South university (MORCS-1)", a mobile robot designed byus that equipped with a 2D laser measurement system to sense the environmentand the proprioceptive sensors such as the odometry, gyroscope to calculate itsdead reckoning, the approach about the four kind uncertainty factors of mobilerobot localization is studied. These researches include that the error analysisand calibration of position sensors is implemented to reduce the measurementnoise, the 3D kinematic model of mobile robot is built to gain the accuratepose in complex terrain, some work on the automatic detection of static ordynamic obstacles based on laser scanner is investigated to eliminate thedynamic influence of the environment and to realize the reliably absoluteposition, and lastly a robust algorithm is presented to involve the incrementalenvironment mapping and self-localization of mobile robot with unknown dataassociation and to improve the self-localization performance of mobile robotunder unknown environment.So, the study in this dissertation focuses on some key points in theuncertainty treatment of mobile robot localization as follows:Combined with the multiple proprioceptive and exteroceptive sensors ofmobile robot MORCS-1, aimed at the drift error of fiber optic gyro as theproprioceptive sensor, the neural network using genetic algorithm as optimaltool is proposed to accomplish the modeling and calibration for temperaturedrift of fiber optic gyro, which can reduce the drift error to the standard outputat constant temperature; and aimed at the noisy disturbance of ranging datafrom the exteroceptive sensor laser scanner, a dynamic adaptive filter isintroduced through the analysis of neighboring ranging data in time and spatialcorrelation to realize the real-time and dynamic filter, which can validly filter the noisy disturbance to meet the requirement of the accurately real-timeobstacle detection in mobile robot navigation.Dead reckoning of mobile robot in complex terrain is analyzed by therigid-body kinematic constraints of mobile robot that is on the basis oflocomotion architecture with the wheeled and rocker-bogie suspension system.At the same time, the kinematic model of mobile robot is obtained using themultiple sensors' information from odometry, fiber optic gyro, tilt sensor, et al.A method of kinematic model integrated with wheel-ground contact angle issuggested to estimate the relative motion trajectory of mobile robot.Experimental results obtained in simulation and with real robot on differentterrains demonstrate that this method is more close to real pose of mobile robotthan to calculate only with the pitch.2D laser scanner is utilized to sense the operating environment of mobilerobot, and the occupancy grids map is imposed to fuse the information of therobot's pose by dead reckoning and the range to obstacles by laser scanner. Anautomatic detection method of static and dynamic obstacles is investigatedbased on 2D laser scanner in non-static environment. The particle filter withthe improved proposal distribution is adopted to track the dynamic obstacles soas to get the localization performance in motion process, and the local mapmatching combined fuzzy logic with maximum likelihood estimation isintroduced to deal with the static obstacles so as to improve theself-localization capability of mobile robot. These methods is verified byexperiments, which shows that it can autonomously divide and detect staticand dynamic obstacles, efficiently track single dynamic obstacle and calibratethe error of dead reckoning.Aimed at the incremental environment mapping and self-localization ofmobile robot with unknown data association, the Rao-Blackwellized particlefilter is improved to get the unite estimation of the pose of mobile robot andthe position of the environmental features. In order to make the right obstacleclassification from the 2D laser scanner, an unsupervised clustering algorithmis presented to realize the feature extraction of obstacles and fuzzy logic isintegrated into incremental data association of obstacles features. Moreover,particle filter for the pose estimation of mobile robot is mended by executing its resampling strategy after the map matching and adapting the resamplingprocess grounded on the effective sample size (ESS). Furthermore, theunscented Kalman filter with the adaptation estimation for the process noise isintroduced into the position evaluation of the environmental features.