LLM-Assisted Emotional Expressive Behavior Generation for Engagement Enhancements in Human-Robot Interaction
Abstract
Social robots are increasingly being integrated into human environments, whether in homes, workplaces, or public spaces. Enhancing user engagement is crucial for effective human-robot interaction, as it determines the quality and sustainability of these interactions across various applications. However, current research on expressive body movements shows limitations in understanding how these movements modulate engagement and attention in real-world contexts. Moreover, these expressive behaviors play an important role in establishing naturalness and effective Human-Robot Interaction (HRI). Addressing this research gap, the present study investigates how expressive robotic behaviors enhance users’ engagement and attention within a storytelling context. Specifically, we designed a novel framework that leverages Large Language Models (LLMs) for generating contextually appropriate robot actions modified by the Laban Effort System for expressiveness. A comprehensive user study (<inline-formula><tex-math notation="LaTeX">$N=289$</tex-math></inline-formula>) was conducted in which users watched the robot storytelling from different perspectives and evaluated the robot’s behavior using engagement and attention questionnaires. The results revealed that Laban-modulated movements yielded statistically significantly higher ratings of engagement (<inline-formula><tex-math notation="LaTeX">$p =. 002$</tex-math></inline-formula>), attention (<inline-formula><tex-math notation="LaTeX">$p =. 005$</tex-math></inline-formula>), anthropomorphism (<inline-formula><tex-math notation="LaTeX">$p =. 003$</tex-math></inline-formula>), and animacy (<inline-formula><tex-math notation="LaTeX">$p < . 001$</tex-math></inline-formula>), while full-body movements yielded statistically significantly higher ratings of animacy (<inline-formula><tex-math notation="LaTeX">$p =. 001$</tex-math></inline-formula>). Our findings provide new insights into how expressive behaviors enhance engagement, with implications for the design of socially interactive robots.
