In this discussion of four spatial representation issues I have identified a number of capabilities that are desirable for a qualitative spatial reasoning system if it is to be used by a robot, and I have also suggested ways of providing those capabilities.
Multiple levels of detail in shape description allow sensory data to be organised at a high level of abstraction, while retaining the information from which the abstract description was constructed. This gives a reasoning system access to all relevant information when the derived high level description is insufficient.
The way in which people make use of the ability to consider complex shape at multiple levels of details can be emulated in two ways: a complete description of each scene can be stored, with index links to appropriate parts of the high-level description. Alternatively, an initial coarse description can be constructed, with the reasoning system directing focus of attention to obtain more detail where necessary. The representation techniques described in the rest of the chapter can support both techniques, by allowing indexing either from coarse to fine levels of detail, or vice versa.
Many human spatial reasoning tasks depend on the ability to consider features or operations in a specific local context. This ability can be used either in concentrating on one portion of a very complex overall shape, or to apply similar operations in similar local contexts, regardless of variations outside the local context.
I describe a specific method for local context shape representation, where an ``imple'' is a portion of overall shape that can be treated as an object on its own. The imple defines local context for size, location, and orientation. Imples also provide a mechanism for description at different levels of detail - the imple context may contain detail that was not appropriate to include at the level of detail description where the imple shape was defined.
The human ability to assign properties to an abstract functional grouping of shape elements provides a number of capabilities that would be useful to a robot spatial reasoning system. These include a record of successful deductions and manipulation strategies, together with a basis for generalisation and identification of previously encountered situation types.
The following representations do not include this facility, but both were designed with consideration to how functional groupings of shape elements might be achieved.
Relative size description is the most basic facility for a qualitative spatial representation. It is the normal mode of operation for people, who seldom need to measure objects before operations can be performed on them. A variety of techniques for relative size description have been proposed, and the two representations discussed below make use of two different techniques - one an axis-relative quantity space, and the other a scene-global size ordering.