Why topological maps are useful for learning in an autonomous agent