Autonomous outdoor navigation and goal finding

The problem of autonomous robotic geocaching, which involves locating a goal object in an unstructured outdoor environment given only its rough GPS position, is an interesting research challenge. The potential benefits of this research include applications to agriculture, search and rescue, surveying, transportation, and interstellar exploration. There are a number of attributes of the autonomous geocaching problem, which encompasses the problems of unstructured outdoor navigation and goal finding, that make it both interesting and challenging. These include the countless outdoor hazards which make obstacle detection difficult; large and interconnected obstacles inherent in outdoor environments (e.g. ditches, dense forest, buildings) which necessitate frequent backtracking and sophisticated path-planning; and the lack of additional information, such as satellite maps, GPS waypoints, or obstacle descriptions, available to the robot. These difficulties have limited previous work in outdoor navigation to the development of systems that rely on structural cues (e.g. road, paths, manually colored obstacles) to aid both obstacle detection and goal finding. This reliance limits these systems to a narrow range of environments.;This thesis presents a fully autonomous robotic system capable of solving the task of geocaching. The final robotic system, named "Kato", was constructed by outfitting a self-balancing Segway robotics platform with various sensors and computing resources. These sensors included a GPS receiver and inertial sensor for position estimation, a laser rangefinder for obstacle detection, and a camera for goal identification. The key challenges in solving the geocaching task include: constructing a map of the terrain as the robot navigates, using this map for planning and following a path to the GPS coordinate, and finally searching the local area for the goal object. The effectiveness of the system was tested in two different outdoor environments, and its performance was compared to that of a human expert teleoperating the robot. Results indicated that, in the test environments, the robotic system was able to navigate to, and detect, the goal object with a high level of dependability. Moreover, path planning and navigational (maximum speed) efficiency and obstacle avoidance was similar when the system was operated autonomously to when it was teleoperated by a human expert. Overall, this work makes strides towards the development of a cost-effective robotic system that can effectively operate in challenging real-world environments.