In this paper we present Region and Gateway (RG) mapping, a novel approach to laser-based 2D line mapping of indoor environments. RG mapping is capable of acquiring very compact, structured, and semantically annotated maps. We present and empirically analyze the method based on map acquisition experiments with autonomous mobile robots. The experiments show that RG mapping drastically compresses the data contained in line scan maps without substantial loss of accuracy.

Cognitive maps are made up of two types of spatial units { landmarks and gateways. While most research has focused upon landmarks, gateways oer an attractive alternative that can open up new possibilities for building spatial representations. One example of this is the crucial role that gateways play in helping to parse environments. Parsing is a crucial step in reasoning about complex environments since it works to break large regions into smaller spaces. Further, since gateways are less directly tied to object recognition they aord a number of advantages for computational models. This paper examines the possibilities of using gateways, structures rst proposed as part of the PLAN architecture, as the basis for spatial modeling. Two systems are reviewed which have implemented this methodology and extensions to the research are proposed, including spatial representations of abstract environments. 1. Introduction It is far from obvious that researchers interested...

Abstract not found

Abstract. Conventional models of navigation commonly assume a navigation agent’s location can be precisely determined. This paper examines the more general case, where an agent’s actual location cannot be precisely determined. This paper develops a formal model of navigation under imprecision using a graph. Two key strategies for dealing with imprecision are identified and defined: contingency and refinement. A contingency strategy aims to find an instruction sequence that maximizes an agent’s chances of reaching its destination. A refinement strategy aims to use knowledge gained as an agent moves through the network to disambiguate location. Examples of both strategies are empirically tested using a simulation with computerized navigation agents moving through a road network at different levels of locational imprecision. The results of the simulation indicate that both the strategies, contingency and refinement, applied individually can produce significant improvements in navigation performance under imprecision, at least at relatively fine granularities. Using both strategies in concert produced significant improvements in performance across all granularities. 1.