Adaptation of Fuzzy Cognitive Maps for Navigation Purposes by Migration Algorithms

Abstract

Fuzzy Cognitive Maps (FCM) represent not only a user-friendly knowledge representation but also a convenient means for simulation of dynamic systems and decision-making support. Concerning the nature of robotic systems FCM seem to be convenient in using mainly on upper decision levels. However, FCM strike on problems of their design. Beside manual approach, which is limited by the number of nodes and their connections, various adaptation methods have been proposed. This paper gives a short summary of these methods dividing them into Hebbian-based and evolutionary-based approaches. Further, it presents a new adaptation of the so-called Self-Organizing Migration Algorithms (SOMA) for purposes of FCM design, which is compared also to other methods like particle swarm optimization, simulated annealing, active and nonlinear Hebbian learning on experiments with catching targets for future purposes of robotic soccer. Obtained results are compared where advantages of the proposed method are apparent and in the conclusions their properties are summarized. Besides, a new modification of FCM with active inputs is presented that is able to receive data from sensors in each time step.