Simultaneous localization and mapping for navigation in realistic environments

Abstract

Navigating autonomously in a domestic environment is a problem that has attracted a great deal of interest in mobile robotics. A robotic system that operates in ordinary furnished rooms without the need of an engineered environment has many different applications such as service, cleaning and surveillance tasks or simply entertainment. Robotic systems that use artificial landmarks or pre-stored maps of the environment are available today. However, these systems are not very flexible. The user must in fact supply a map of the environment, which can be interpreted by the system. This thesis deals with the problem of Simultaneous Localization and Mapping (SLAM). The mobile robot builds a map of an unexplored environment while simultaneously using this map to localize itself. The feature based approach used in this thesis utilizes the Extended Kalman Filter (EKF) machinery to estimate the pose of the robot and the location of the features. This approach is referred to as stochastic mapping. Point features in the environment are robustly extracted from sonar data using triangulation techniques. In addition, this thesis explores a method for recovering from the most common mode of failure of the stochastic mapping approach. This method allows the EKF algorithm to continue in a consistent manner after a failure has been detected. Finally, the thesis presents a method for achieving more accurate navigation by using the architectural properties of most domestic environments. This method drastically improves navigation, when the stochastic mapping algorithm can not be used due to poor quality sensor data. All the algorithms presented in this thesis have been tested and verified in real world experiments. Keywords: mobile robots, sensor fusions, sonars, odometry, SLAM, Kalman filter, mapping, localization, navigation.