Spatial understanding and temporal correlation for a mobile robot

In this article we discuss a network model, which simulates functionallysome of the features intrinsic to human navigation and the merits ofincorporating such features on a behavior-based robot. Specifically thepaper deals with implementing a memory based reasoning strategy duringreal-time navigation of a mobile robot. For the purpose ofimplementation, memory has been identified with an ability to cognizethe local environment or scenario, classify such scenarios in terms ofpreviously learned primitives or landmarks, remember and recollect suchprimitives at later instants and correlate over time similar experiencesof scenarios. Such memory based reasoning enhances the robot'snavigation capabilities through intelligent decisions due to spatialunderstanding, scene recollection abilities by remembrance and detectinglocal minimum traps through place recognition. A double layeredspatio-temporal classification scheme consisting of a fuzzy rule-basedspatial classifier and a temporal classifier based on self-organizingmap and ART networks are adopted for this purpose. The classifiernetwork reduces the robot's experience of its environment consisting ofa stream of sensor patterns into weight vectors that signify aparticular landmark. An extension of the network architecture is alsointroduced to cognize the presence of dynamic obstacles amidststationary ones.