Learning variable-resolution maps for navigation in dynamic worlds

Map building is an important component to enable autonomous mobile robots to navigate in complex unfamiliar environments. This paper introduces a new method for learning multiresolution maps for navigation in unknown and dynamic worlds. It extends our navigation architecture that integrates a multiresolution grid world model and fuzzy ART based world model. The fuzzy ART world model is composed of a set of rectangular geometric primitives, or features. In a companion paper we introduce a new method for updating the fuzzy ART model in dynamic worlds. In this paper we describe our new navigation architecture that, by integrating the new proposed map building method for dynamic worlds, is able to dynamically not only increase but also decrease local resolution according to variations in the local clutter and complexity of the world. With the new overall navigation architecture the mobile robot is able to cope with, and navigate, in changing worlds. The paper presents experimental results obtained with a Nomad 200 mobile robot that demonstrates the effectiveness of the proposed methods.