Layered Knowledge Architecture For Navigation-Oriented Environment Representation

This paper describes a layered architecture representing knowledge to be used for navigation in environments where distinctive places are present. Each knowledge layer defines an abstraction of the environment aimed at efficiently supporting the execution of specific navigation tasks. Layers are structured in a taxonomy, each layer being derived from another by applying one of the three abstraction primitives: classification, generalization, aggregation; in particular, aggregation enables partitioning of the environment map into clusters. The validity of the architecture proposed is discussed with reference to different application fields, including autonomous robots and vehicle navigation systems. Some issues related to hierarchical path planning on layered knowledge are then addressed; a general framework for a more flexible formulation of path-planning problems and a technique for decomposing problems on the different layers are proposed. An efficient algorithm to solve a specific navigation problem, named "travelling agent", is presented; the algorithm uses a heuristic criterion to expand, at each step, a path computed within a layer into a path at the next layer down. Index-terms abstraction, clustering, hierarchical planning, knowledge representation, path planning