Hybrid knowledge representation applied to the learning of the shared attention

Abstract

Sociable robots are embodied agents that are part of a heterogeneous society of robots and humans. They are able to recognize human beings and each other, and engage in social interactions. The use of a robotic architecture may strongly reduce the time and effort required to construct a sociable robot. However, a robotic architecture for sociable robots must have structures and mechanisms to allow social interaction, behavior control and learning from environment. In this article, a new hybrid knowledge representation is proposed and integrated to our robotic architecture inspired on Behavior Analysis. This new hybrid knowledge representation enables incremental learning and knowledge generalization by incorporating an ART2 neural network combined with a relational presentation of first order. The new representation has been evaluated in the context of the learning of the shared attention and the results obtained show that it is a very promising approach.