Deep Learning Approaches for User Engagement Detection in Human-Robot Interaction: A Scoping Review

The increasing use of social assistive robots (SARs) has sparked researcher interest to investigate user engagement to enhance SAR interactive capabilities. Engagement in Human-Robot Interaction (HRI) aims to benefit users during interactions. Diverse interpretations of engagement have led to various metrics for its measurement and detection. Despite numerous algorithmic approaches for detecting user engagement, Deep Learning (DL) algorithms have become prominent in HRI engagement detection. However, there is a lack of comprehensive reviews on DL methods for engagement detection in HRI. This scoping review summarizes a decade of DL applications in HRI engagement detection, highlighting key findings and gaps including the need for context-specific datasets, understanding temporal dynamics, and exploring non-social robots. Moreover, this review focuses on employed DL algorithms, sensory inputs, ground truths, robots, and datasets. This review serves as a valuable reference for HRI researchers aiming to improve user engagement detection strategies.