Next: Contributions Up: Conclusions and Future Work Previous: Conclusions and Future Work
The need for autonomous robots has been rapidly increasing in the last years. There are many areas in which these robots are used, ranging from ``service robots'', such as museum guides or transportation robots in factories, to robots used for tasks to be performed in inaccessible environments, such as planetary exploration, hazardous material handling and rescue missions.
Usually, service robots operate in indoor structured environments. The problem of navigating through indoor environments has been the focus of robotics research during many years, and many successful results have been achieved. Usually, the map of the environment is given a priori (either a detailed metric map or a topological one, showing the spatial relationship among different places of the environment), or, when it is not given, there is an initial phase for learning the map. Once it is learned, the robot repeatedly performs the task in this environment. Examples of such robots are those performing delivery tasks in office environments or guiding tours in museums [67,9].
On the other hand, inaccessible environments are usually unknown and unstructured (as is the case in most outdoor environments), which pose a more difficult problem. The lack of structure of such environments makes the map building very difficult. Moreover, the large scale of these environments also adds to the difficulty of mapping and navigation tasks. These characteristics make it impossible to apply the approaches used in indoor structured environments. Although there has been also a lot of research on navigation in unstructured environments, it is still an open problem.
This PhD thesis has focused on this latter problem, that is, on navigating in unknown unstructured environments. The research was part of a robotics project whose goal is to have a completely autonomous robot capable of navigating in outdoor unknown environments. A human operator selects a target using the visual information received from the robot's camera, and the robot has to reach it without any further intervention of the operator. Navigating to a target is a fundamental task of any mobile robot, whatever its mission is (be it grasping objects, analyzing them, looking for something, etc.) The task to be performed once the target has been reached is outside the scope of the project and this thesis.
A first milestone of the project was to develop a navigation system for indoor unknown unstructured environments. The reason for starting with indoor environments was that the development of robust vision systems for outdoor environments is still an open and very difficult problem in the field of computer vision. Therefore, since the vision system was not the focus of our research, we decided to start experimenting indoors, for which vision systems are much easier to develop. Moreover, we designed the landmarks so that we could easily change their location, thus, permitting us to configure scenarios of different complexity.
This thesis has reported the research carried out in order to accomplish this first milestone. For achieving it, we have combined landmark-based navigation, fuzzy distance and angle representations and multiagent coordination based on a bidding mechanism. The objective of our research was to have a robust navigation system with orientation sense for unknown unstructured environments using visual information.
© 2003 Dídac Busquets