A Navigation System for Indoor/Outdoor Environments with an Unmanned Ground Vehicle (UGV)

Abstract

This thesis presents an approach for solving the global navigation problem of wheeled mobile robots. The presented solution for outdoor navigation uses Extended Kalman Filter (EKF) to estimate the robot location based on the measurements from Global Positioning System (GPS), inertial measurement unit (IMU) and wheel encoders. For indoor navigation (where GPS signals are blocked) another probabilistic approach, based on Monte Carlo Localization (MCL), is used for localization. This algorithm utilizes the map of the environment to estimate the posterior of the robot using the depth measurements from a Kinect sensor. The output from the Kinect sensor is processed to imitate the output of a 2D laser scanner by projecting the points from a thin horizontal strip of pixels in the image plane to the corresponding real world 3D coordinates using the pin-hole camera model. Two different controllers based on Dynamic Feedback Linearization (DFL) and Input-Output State Feedback Linearization (I-O SFL) have been analyzed, simulated and compared. Based on the thesis objective and the simulated results, the I-O SFL method was chosen for solving the trajectory tracking problem. A set of test experiments was conducted to evaluate the performance of the proposed system in outdoor, indoor and a combination of both environments. The results show that the robot can successfully navigate through the way-points with a great accuracy in indoor environments, while the accuracy in outdoor environments is within the 3m position accuracy of the GPS.