Spatial abstraction empowers complex agent control processes. We propose a formal definition of spatial abstraction and classify it by its three facets, namely aspectualization, coarsening, and conceptual classification. Their characteristics are essentially shaped by the representation on which abstraction is performed. We argue for the use of so-called aspectualizable representations which enable knowledge transfer in agent control tasks. In a case study we demonstrate that aspectualizable spatial knowledge learned in a simplified simulation empowers strategy transfer to a real robotics platform.