View-based cognitive map learning by an autonomous robot

This paper presents a view based approach to map learning and nav igation in mazes By means of graph theory we have shown that the view graph is a su cient representation for map behaviour such as path planning A neural network for unsupervised learning of the view graph from sequences of views is constructed We use a modi ed Kohonen learning rule that transforms temporal sequence rather than fea tural similarity into connectedness In the main part of the paper we present a robot implementation of the scheme The results show that the proposed network is able to support map behaviour in simple environ ments Introduction The view graph representation A cognitive map is a neural mechanism supporting navigation and orientation tasks much as a real map of the environment Sch olkopf and Mallot presented a mechanism for the learning of a cognitive map of a maze from the sequence of local views encountered when exploring the maze In this approach the topological structure of a maze is represented as a graph where the places are the nodes and the directional corridors are the edges see Fig a The explorative sequence of encountered views corresponds to the sequence of directed corridors passed along the way through the maze Instead of reconstructing the place graph explicitly e g Kuipers and Byun we consider an intermediate representation which is called the view graph i e a graph whose nodes represent the encountered views and whose directed links represent the temporal coherence of views Fig b If one assumes that each corridor corresponds to exactly one view and all views are distinguish able the view graph can be shown to contain all of the information required to reconstruct the place graph Sch olkopf and Mallot In the present paper