Cognitive Map Construction and Use: A Parallel Distributed Processing Approach

Abstract The Connectionist Navigational Map (CNM) is a parallel distributed processing architecture for the learning and use of robot spatial maps. It is shown here how a robot can, using a recurrent network (the CNM predictive map ), learn a model of its environment that allows it to predict what sensations it would have if it were to move in a particular way. It is shown how this predictive ability can be used (via the CNM orienting system ) to enable the robot to determine its current location. This ability, in turn, can be used, when given a desired sensation, to generate sequences of goal states that provide a route to a place with the desired sensory properties. This sequence is given to the CNM's inverse model , which in turn generates a sequence of actions that effects the desired state transitions, thus providing a sort of “content-addressable” planning capability. Finally, the theoretical motivation behind this work is discussed.