Spatial Reasoning with Uncertain Data Using Stochastic Relaxation

Abstract

Spatial Reasoning often uses sensor data as input. Since sensor data normally have distortions problems of uncertainty and possible contradictions arise. Classical AI inference techniques have diÆculties in dealing with uncertainty and contradictions. Thus our approch is to use soft computing techniques for our domain. We augmented the simulated annealing approch by composition tables as ternary constraints and relation sets as variables. Some constraints model physical laws which have strong condence. Others measure data with much lower condence. Therefore we also have to integrate soft and hard constraints. Keywords: approximate reasoning, hybrid systems, spatial reasoning 1 Introduction Sensor data are often used as input by spatial reasoning. Examples are the analysis of satelite images in geographical information systems, the speech recognition in natural language interfaces and landmark identication in robotic navigation. However, sensor data may include contradictions. Cla...