Practical Navigation: Autonomous Behavior on Inexpensive Robots

Abstract

While the problem of fully autonomous localization and navigation has been studied at length in theoretical terms and implemented in robots with complex suites of sensors, the feasibility of autonomous robotics in less robust systems is a largely under emphasized problem. However, as inexpensive robots become more available, understanding their viability as autonomous, position oriented systems becomes more pressing. This thesis explores the practical implementation of a fully autonomous system in an typical building environment using the humble Turtlebot platform. In our target environment, we have access to blueprints, but limited control over moveable and dynamic obstacles (e.g., furniture and people). With this in mind, we work from the basic building block of autonomous navigation: localization and local navigation. We will use a floorplan and landmark based localization system combined with metric path planning for global navigation and bug-based local navigation. Each component is thoroughly documented and tested before ultimately being integrated into a complete autonomous system. Good results are achieved using these methods, although the mechanical imperfections of the Turtlebot make periods of uncorrected local navigation prone to drift.