Intention estimation and recommendation system based on attention sharing

Abstract

In human-agent interactions, attention sharing plays a key role in understanding other's intention without explicit verbal explanation. Deep learning algorithms are recently used to model these interactions in a complex real world environment. In this paper we propose a deep learning based intention estimation and recommendation system by understanding humans attention based on their gestures. Action-object affordances are modeled using stacked auto-encoder, which represents the relationships between actions and objects. Intention estimation and object recommendation system according to human intention is implemented based on an affordance model. Experimental result demonstrates meaningful intention estimation and recommendation performance in the real-world scenarios. © Springer-Verlag 2013.