In this paper, we present a method capable of identifying and locating objects with known three-dimensional models in a single perspective image. The identity and location of an object are obtained through a procedure which we refer to as a knowledge-directed search. The search encodes declarative and procedural knowledge in a rule network designed to ensure that only pertinent information is processed. The search is organized as a collection of search activations. Each search activation has a context memory for recording previously inferred or assumed information. As a result, multiple search activations may be used to test various hypotheses of object identity and location.

This paper presents a method of structuring the free space of a roving robot's environment into a set of overlapping convex regions ideally suited to path planning and navigation tasks. The structure of the free space environment is maintained as a hypergraph with each convex region represented by a hyperedge identifying the boundary walls of the region. A new methodology reveals the structure of free space and constructs the hypergraph representation through a directed search for a set of fundamental circits in an abstract graphical representation of the environment geometry.