When a mobile robot performs a task within a given environment, it needs to have some knowledge of that environment to effectively carry out the task. Generally, the environments in which robots operate are unstructured, complex, and changing. Thus, it is crucial to create models of these environments based on the information and observations captured by the robots within them to ensure effective localization. This is the focus of the project proposal.

The main objective we propose is to solve the problem of creating maps of an unknown environment, using the information provided by a vision system installed on the robot exploring the environment. The traditional approach to solving such problems involves extracting local features from scenes and creating metric maps in which the robot's position can be estimated relative to a global reference system, with an associated error. In contrast to this approach, we propose using the global appearance information of scenes to create topological maps, which contain information about the locations that make up the environment and the connectivity relationships between them. These are more recent and computationally efficient alternatives that, however, require in-depth study in tasks of creating robust maps of extensive and dynamically characteristic environments.