Doctor of Philosophy
School of Information Technology and Computer Science
Ratner, Danny, Landmark navigation of the titan 4WD outdoor mobile robot using narrow beam CTFM sonar, Doctor of Philosophy thesis, School of Information Technology and Computer Science, University of Wollongong, 2002. http://ro.uow.edu.au/theses/2015
Navigation is the science (or art) of directing the course of a mobile robot as it traverses the environment (land, sea, or air). A robot that is following a path on a map has to check its location regularly. Its estimate of its location can drift due to odometry and other errors. As the estimate drifts, the error between its actual location and where it thinks it is on the map increases. If this error is not corrected, the robot can collide with an object or fail to dock correctly. A robot must use external sensors to detect landmarks in the environment. From measurement of its range to these landmarks, it can calculate its actual position using triangulation. The accuracy of this method is limited by the sensor's ability to identify objects correctly.
Piloting is a navigation strategy that uses known landmarks. These landmarks are used sequentially to find the way to the goal. The navigator must be familiar with the area, and know which landmarks to look for. A landmark is a feature in the environment, whose position can be sensed, that is close enough to the desired path that its direction varies significantly as the navigator moves along the path.
Titan's navigation concept is inspired by the flying experience of light aircraft in long cross-country trips using Visual Flight Rules (VFR) to navigate. It is composed of a blend of three basic navigation modes: landmark navigation legs, bridging navigation legs and path edge following. Light aircraft usually fly in good visibility between a set of predefined landmarks. The pilot uses the recognition of landmarks to assess the current localisation error and calculate the needed correction to the navigation. Landmarks are prominent geographical features (river bends, lakes etc.) or structures (power stations, bridges, etc.).
Most of the time the landmarks are isolated so the flight along the leg is without reference to the scenery below, but sometimes it is convenient to stick to a continuous landmark like a road or coast line. In the case of bad visibility, the pilot will fly an IFR (Instrument Flying Rules) leg by watching the aircraft instruments, without being able to use landmark recognition.
Plants are naturally occurring landmarks. They can be detected and recognized using Continuous Transmission Frequency Modulated (CTFM) ultrasonic sensing. Landmarks can be grouped in four classes based on their geometric complexity and their continuity. The landmark class determines the sensor motion strategy, the recognition feature set and the navigation strategy. This thesis reports on research into landmark recognition for each of these landmark classes.
Robotics involves the integration of sensing systems, motion control systems and manipulation systems using software. Traditionally, this software includes a real-time operating system supporting multiple processes, all written in C/C++. An alternate approach to developing a software architecture for machine perception and robot control is to use off the self high-level tools.
In this thesis, we discuss the use of software tools for the design, modeling and control of a 4 wheel-drive mobile robot. AutoCAD is used for kinematic analysis. Simulink is used for modeling the robot dynamics. Matlab is used for designing the control and recognition system. Lab VIEW is used for programming the robot control and machine perception systems. Support for the tools was obtained by interaction with the technical support groups of the relevant suppliers over the Internet. The research is conducted without the need to write a single C/C++ line of code.
The result of this research is a mobile robot that can navigate in outdoor environments using both continuous and discontinuous landmarks. The landmarks are detected with a phased array CTFM ultrasonic sensor. The robot is taught a map by manually driving it along the desired path and recording landmark and odometry information. It can then retrace the path, using the sensing of landmarks to re-localise goes, so that it robustly follows the taught path.