Learning and Adaptation of Cognitive Distance in Directed Multi-Agent

Abstract

This paper deals with an embodied cognition and decision making in directed multi-agent systems (DMAS), in which a well-organized team play takes more important role than conventional MAS. Due to a strong physical constrain, the directivity in the agent behavior has to be taken into account. We assume that a sense of distance is also affected by the physical constraint, hence we proposed the concept of Subjective Distance, which is perceived subjectively for a directed agent. Unlike general definition of mathematical distance, the subjective distance between agents is supposed to be asymmetry. Therefore, exploiting the asymmetry of the subjective distance is expected to facilitate a well organized team play. Because the directivity is perceived indirectly rather than given as a priori given property, the directed agent is required to learn and adapt its directivity parameter. This paper describes a learning algorithm of the directivity parameter so that the subjective distance is evaluated in DMAS.