The visionary goal of an easy to use service robot implies some key features like spatial cognition, speech understanding and object recognition. Therefore such a system needs techniques to identify objects in scenes, i.e. to assign the natural category (e.g. "door", "chair", "table") to new objects based on their prototypical geometry.

Our approach uses 2{1 \over 2} D laser range data to recognize basic objects like chairs or tables within an office environment. It is based on the concept of affordances; established on the work about form and function we identify certain geometries that lead to certain functions and therefore allow their identification. Our approach currently is restricted to basic objects but not limited to a special form. This is achieved by spatial abstraction where we assign the data to three layers. In identifying components in these layers of altitude, we reconstruct the basic form of the object, conclude its function and finally determine the object.