Problem-solving approach to the localization problem
Bonnie Kathleen Holte Bennett · 1992
This research describes a approach to the localization problem. Localization is the process by which an agent determines its current location with respect to a map. The term problem-solving implies a computational technique based on logical representational and control steps. In this research, these steps are derived from observing experts solving localization problems. The objective is not specifically to simulate human expertise, but rather to apply its techniques where appropriate for computational systems. This document analyzes the localization problem. Localization requires that correspondences must be established between map and view features. The problem is sufficiently complex that a simple brute force search is too computationally expensive to be useful. An approach is required to focus the search on promising solutions. Past work has focused on low-level processing of image inputs, or has relied on an estimate of the current location; however, this research focuses on the high-level reasoning components that can provide heuristic control to solving the localization problem without an a priori estimate of location. We will argue that using the heuristics of selectively grouping information and concurrently working at multiple levels of detail will solve the high-level reasoning requirements of the localization problem. Further, we describe a novel approach to solving localization by applying a methodology. In doing this, we describe a model for solving the problem and a system built on that model called Localization Control And logic Expert (LOCALE). LOCALE is a demonstration of concept for the approach and the model. The results of this work represent the first successful solution to high-level control aspects of the localization problem. The approach uses a heuristic strategy to deal with potentially combinatorially explosive data in a reasonable manner.