A topological approach to online autonomous map building for mobile robot navigation

This thesis presents an online autonomous mobile robot exploration and navigation strategy. An appropriate environmental representation is an essential part of an efficient navigation system. We choose a topological map representation, where the world is represented by a set of omnidirectional images captured at each node with edges joining the nodes. Topological maps are memory-efficient and enable fast and simple path planning towards a specified goal. Using a laser range finder and an omnidirectional camera, an online topological representation of the environment is developed; although the navigation process relies only on the omnidirectional camera. We choose to use an omnidirectional camera, because it gives a 360 horizontal field-of-view and offers other advantages, such as increased robustness to occlusion, rich information content, etc. A view classifier based on global image comparison technique is used in order to avoid the possibility of creating a node in the same or nearby location where another node was created in the topological map. -- A robot navigation system is presented which is based on visual information only. The visual homing mechanism is used to move the robot from one node to another in the topological map. Visual homing can be defined as the ability to return to a goal location by performing some kind of matching between the image taken while at the goal and the current view image. Path planning algorithm is implemented for successful vision-based navigation. All the experiments are done in an office environment using a Pioneer 3AT mobile robot. The topological map is built in real time on board the robot, thus making the system autonomous