A Semi-Centralized Layered Route Planner for Navigation Systems

Abstract

This paper describes a new route planning model with three layers (two symbolic layers and one subsumption layer) for a mobile robot and the result of an experiment in computer simulation. Each layer consists of a map-generator and a route-planner, and a plan monitor coordinates three layers. These three layers are classified according to the term of generated plans in a top-down design, in which a lower planner refines a part of the route generated by the higher in sequence. The lowest planner decides a direction on the spot to get closer to a goal, while the others generate or refine to a goal based on A algorithm. Though higher layers take control of lowers by giving representation (A route on a map), each module is asynchronous and drives replanning by comparing the representation with situation. For example, a map-generator will suggest re-planning, when the environment change of its field is more than a given parameter. By giving two layers to symbolic reasoning, the load of symbolic reasoning in dynamic planning and wasteful planning cost in repairing plans can be reduced. In addition, the plan monitor takes a role of small repairing plans without driving symbolic reasoning. The model is implemented on UNIX workstations.