Error Correction In Large-Scale Cognitive Maps

Abstract

ABSTRACT A mobile robot that maintains a dynamic cognitive map willoften find that the information in the map is contradicted by hisperceptions, and is therefore incorrect. Such errors may be the result of an earlier misperception, an erroneous matching, an erroneous default inference, computational errors, a change in theworld over time, or an erroneous previous error correction. Due tothe complexity of inference in forming cognitive maps, domain - independent strategies for error correction, such as data-dependencies or conditional probabilities, are not sufficient by themselves to give a robust error correction scheme. Rather, domain -specific techniques and heuristics must be applied. We dis-cuss some of the basic issues involved in detecting, diagnosing and correcting errors in the cognitive map. We also discuss how arobot may decide whether to take actions in order to gather relevant information.1. INTRODUCTIONMost animals, over the course of their lifetime, move about in an environ-ment that is considerably larger than the range of their sensors and is notunder their direct control, but is stable enough that many features remain thesame between one visit to a place and the next. Such an animal, or a mobilerobot with sensors that operates under similar circumstances, stands to gainfrom learning and remembering the geographic characteristics of its environ-ment. A dynamic cognitive map is a knowledge structure that supports suchlearning and remembering. There have been numerous studies' proposingknowledge structures for dynamic cognitive maps for use in AI systems.It is not possible, in general, to design a cognitive mapping system thatis both powerful enough to be useful to a robot in a rich environment andalso secure enough to be guaranteed correct. In almost any uncontrolledenvironment, there are possible circumstances in which the cognitive map-ping system may perform an operation which, though reasonable, is in factmistaken and results in the cognitive map being incorrect. A robust cognitivemapping system must therefore have the capacity to detect and deal witherrors. Previous cognitive mapping systems that have avoided doing errorcorrection are necessarily fragile. A number of systems47 -10 have addressedthe problem of correcting perceptual errors, but these have generally usedoverly simple models that are appropriate only in very restricted environ-ments. In this paper, we give a preliminary discussion of the issues that arisein error correction in a broad range of cognitive mapping systems. We donot, however, give any complete algorithms for any specific cognitive map-ping architecture.