Human attention estimation by a domestic service robot using upper body skeletal information

Abstract

Human-robot interaction is widely used for service applications in serving elderly and disabled population. A humanlike behavior is expected from these robots to achieve naturalness in interaction. Humanlike behavior is important before as well as after initiation of interaction with a human user. Basic cognitive skills are required for analyzing such a scenario involving humans. Many existing robots cannot interpret attention of its user before engaging with him. This paper presents a model to interpret the level of attention of the user when robot approaches him. The system extracts upper body information to approximate the level of attention of its user. This has separate functional units to extract and analyze information, estimate the attention level and finally to take decisions regarding interaction initiation with its user. It involves fuzzy evaluation of important parameters which affect the attention of a human. These parameters involve behavior of upper body of the person and using these, robot estimates the level of attention of its user without directly interacting. This model is implemented and tested in a domestic environment with users of a broad age gap. Implementation of this model and results of the experiment are presented.