Spatial reasoning, as a very basic human ability, has always been a priority in the investigation of AI reasoning systems. Hayes identified spatial reasoning as a particularly important component of the Naive Physics project, and spatial domains have often been chosen for experimental Qualitative Physics systems. This consensus on the importance of qualitative spatial reasoning has not, however, resulted in many qualitative systems that have general purpose spatial reasoning abilities.
The issue that most clearly separates the requirements of spatial reasoning from previous research in qualitative reasoning is the representation of state. Qualitative physics systems reason about their domains by identifying and forecasting changes of state in a network of discrete devices. It is difficult to describe general motion in terms of change of state in a network, because free space is continuous, not discrete. Previous approaches to state-based qualitative spatial reasoning have involved the division of free space into a network of ``discrete'' regions, abstraction of physical devices so that they can be represented simply as linked nodes, or simplification of motion to mean only transitions between previously identified contact states.
Existing qualitative spatial reasoning systems have usually incorporated a preprocessing stage which uses conventional numeric techniques to determine possible qualitative motion states. The qualitative part of the system then analyses the dynamics of this qualitative description. The disadvantage of this technique is that the qualitative part of the system does not have direct access to a description of the scene geometry. The range of problems addressed by qualitative physics shows the consequences of using state information that does not include geometry - they can easily represent the processes involved in a mechanism (energy transfer, change of state), but they cannot reason about the simpler concepts of relative motion and constraint.
The partial distance ordering/extended polygon boundary representation of shape and position is a purely qualitative representation - it includes no quantitative information at all. It is at least as powerful for spatial description as the representations that have previously been used by qualitative reasoning systems in spatial domains. To illustrate this, the following section discusses three well-known domains for qualitative spatial reasoning, and shows how the PDO/EPB representation could be used in those domains.