A mobile robot needs a model of its environment to perform planned tasks. Abstraction-based (hierarchical) arrangements of information allows the robot to plan its tasks efficiently. In this paper, a model with multiple hierarchies of abstraction is applied to mobile robots. It provides better adaptability to a wider range of environments, tasks, and agents than single-abstraction models. A system is also presented which automatically constructs such a model. It is inspired on the idea that the best model for an agent is the one that allows it to operate most efficiently. Some example applications to mobile robots illustrate these ideas.