Mobile Robot Localization And Map Building Based On Multi-Sensor Data Fusion

With the fast development of new technology in intelligent control, computer science, networking, binoics and artificial intelligence, mobile robot has become the focus in the field of robotics and automation. Localization and map building are two essential tasks for autonomous mobile robot navigation. The intension of this paper is to introduce our research on nonholonomic mobile robot localization and map building in the indoor environment with uncertainty information.The SmartROB-2 mobile robots used in our research have a multi-sensor setup where two LMS 200 laser range finders provide accurate position information about the surrounding objects in 360° field of view, monocular vision data obtained from CCD camera offer partially redundant environment information, omni directional tactile bumper and encoders used for odometry are also very helpful in autonomous mobile localization and map building.This paper not only introduces the nonholonomic robot's speed and position controller, but also analyzes various infections on different sensors firstly. It presents a characterization study of the LMS 200 laser scanner and shows the advantages of laser scanner in mobile robot research. Uncertainty model in range measurement is built based on the referenced experiment results. Then it points out the limitation in dead-reckoning and gives out the state covariance matrix. In addition, the camera system is calibrated and compensated against radial distortion successfully.Feature extracting, merging and fusion are decisive factors for mobile robot localization and map building. All of the localization approaches in this paper are feature based instead of directly using the raw data from on-board sensors. Due to the variant sensor modeling for laser range finder and CCD camera, weighted least square fitting and non-local maximum suppression algorithm are used to extract certain 2-D horizontal environmental features and vertical edges respectively. After the further matching between horizontal environmental features and vertical edges, the robot can use the high-level feature as natural landmark in localization and navigation.The representation of environment is crucial for localization and map building. In this paper, research on mobile robot localization with priori map can be classified as extend Kalman filter-based approach, probabilistic localization approach and hybrid metric-topological localization. In order to implement mobile robot indoor navigation without a priori map, this paper discusses methods in mobile robot indoor simultaneous localization and mapping using laser scanning and monocular vision. Based on the real mobile robot platform, we present an approach to complete EKF localization and metric map building simultaneously based on the result of lines merging and feature fusion. Experiment results implemented in the SmartROB-2 mobile robot and further experiment data analysis show the method's validity, robustness and practicability.