The most important part of a space filling representation is the way in which the abutment of occupied space regions against empty regions is described. The refinements of space filling methods in the previous section all involve ways of economically improving accuracy along this boundary. A common technique in shape representation for robots is therefore to describe this boundary explicitly, rather than as an implicit line between occupied and unoccupied regions.
Explicit boundary representation provides important benefits both in pure vision systems, and in systems where a robot arm is controlled using a visually acquired shape representation. An example of the former, developed for industrial applications, is quality control and sorting of objects lying flat on a conveyor belt, where boundary shape is the criterion used for sorting. Typical examples are quality control which checks that biscuits are round (no pieces broken off), or a sorting machine for fancy chocolates [Cro82], which stores mathematical models of the curved boundaries of each type of chocolate.
The boundaries of a workpiece are very important in controlling a robot arm, since it interacts with the workpiece only along those boundaries. A good example of this is Trevylan's robot sheep shearer [TKO82], which constructs a mathematical model of the boundary of each sheep. This model is used to guide the shearing arm over the sheep's skin.
Both recognition and robot handling using a simple boundary description are described as goals of a system which, among other tasks, identifies and handles kiwifruit [FA86]. This system describes the ``silhouette polygon'' of the kiwifruit in terms of a series of vectors around the boundary.
In each of the above cases, the shape boundary is the only information which is relevant to the robot task, so the best shape representation is to make explicit those parts of the boundary which the robot must operate with.