Very few attempts have been made (to date) to perform robot reasoning tasks using qualitative methods. Robot controllers receive information about the world in a numeric form from sensors, and they must supply precise numeric information in order to position and control actuators. Most established algorithms for analysing either sensory information or proposed motion are purely numeric, and most robots are programmed numerically (although the numeric nature of the program may be hidden from the programmer by the use of symbolic programming languages, or data capture devices such as ``teach pendants'').
One of the few examples of an application in which a qualitative representation was used in a robot context is Burger and Bhanu's system for qualitative motion understanding [BB87]. The qualitative representation used here was the output format of a sensory system which filtered information gathered by a vehicle moving over outdoor landscapes. The use of the term ``qualitative'' here implied mainly that the level of detail in the description is similar to that used by people - the system did not carry out qualitative reasoning of the type performed by qualitative physics systems.
There are no qualitative robot reasoning systems which can be used as a direct basis for comparison when evaluating the application of the PDO/EPB representation to robot reasoning, but the cases on which the PDO/EPB representation was tested (described in the last chapter) are very similar to some cases used to test experimental robot planning systems. One example is the disassembly planner described by Sedas and Talukdar [ST87]. This system plans how to remove components of an assembly from a constrained position into free space - essentially the same as the path planning problem formulation described in the last chapter. The PDO/EPB based system essentially solves the same problem, but does so without making use of numeric information.
There are three major advantages that the qualitative spatial reasoning methods presented in this thesis might provide when applied to robot problems. These are: the ability to operate with incomplete information, the ability to degrade gracefully, and the ability to support spatial reasoning at a level which is easily related to human problem solving. The following three sections briefly discuss each of these.