Space and Shape Representation for Robots

All robot systems operate in the physical world, and a robot controller therefore must include some form of spatial representation, whether it is an explicit description of object shapes and locations, or whether it is implicit in a sequence of motions through space. The requirements of robot spatial representations vary considerably in different applications, however - the sophistication of spatial representations must increase as the robot controller is required to perform more sophisticated tasks. The following list presents the types of spatial representation facilities that are necessary to carry out different levels of robot programming and control:

• Robot level programming does not assume that the robot has any explicit knowledge about its surroundings. The shape of the objects that the robot is operating on is, however, implicit (to some extent) in the motions that the robot makes.
• Vision systems (those which perform object recognition, or scene analysis) must have some explicit spatial representation, since all visual processing operates on a two dimensional projection of the physical subject.
• ``Hand-eye'' systems must be able to relate the description of an object as it appears in the visual field to operations that the robot will carry out in its workspace. This requires a spatial representation which is more versatile than for vision alone.
• Systems in which object descriptions are directly created by a programmer require that the description can be readily derived from the programmer's own ``spatial representation'' - his concepts of space and shape. At the same time, the robot controller must be capable of relating the programmer's description both to features in the visual field, and to actual movements. Representation facilities like these are a requirement of task level programming.
• Where a robot is reasoning for itself in an environment that may contain unknown objects, or unfamiliar arrangements of objects, it must be able to use sensory data to construct a useful spatial representation of its surroundings, so that it can plan and operate in those surroundings.
• ``Robots'' that operate in a simulated world are of course particularly dependent on the form in which the simulated world is represented, since they do not interact with anything other than the representation.

The remainder of this section discusses several approaches which have been taken to space and shape representation in robotic systems. It includes discussion of techniques that have been used in most of the above categories of task, so that the capabilities of various general methods can be compared.