An approach to geometric reasoning in robotics

Anticipating the use of knowledge-based expert systems for robot task planning, a geometric reasoning is envisioned for the representation and manipulation of geometric information in a manner uniform with other knowledge. The general requirements for such a reasoner are outlined and a proposed architecture is described, built on four concepts: a class structure of spatial sets, features as subsets of those spatial sets, geometric abstractions as partial representations of features, and constraints as a mechanism for reasoning. An example implementation is built in an object-oriented environment and uses an underlying solids modeler, transparent to the end user