Exploring the Dynamics of Human-Robot Interaction: Robot Error, Sentiment Analysis, and Politeness

Abstract

Abstract As robots become more integrated into daily life, understanding natural human communication is essential for developing user-friendly human-robot interaction (HRI) systems. While prior research focuses on robot-initiated behaviors, little is known about how user-initiated politeness, sentiment, and communication patterns influence perceptions of robots in real-world tasks. This study addresses this gap by analyzing interactions between 35 participants and Alex, a mobile manipulator that understands unscripted speech, during a collaborative cooking task. The study explores four hypotheses: 1) The average number of words per command used for instructing the robot is affected by how the person perceives the robot’s interaction., (2) The robot errors affect the number of words used as well as how the interaction is perceived, (3) The politeness level of the individual expressed during human-robot interaction is affected by the perceived usefulness of the robot, and (4) The sentiment (negative, neutral, or positive) of the individual expressed during human-robot interaction is affected by the perceived ease of use, safety, and trust in their interactions. Kendall’s Tau correlation was used for analysis. Results showed a negative correlation between robot errors and the minimum number of words used ( $$\tau $$ = − 0.39, p = 0.0027), and lower politeness when users felt tasks were completed successfully ( $$\tau $$ = − 0.36, p = 0.0238). Sentiment analysis revealed that slowing speech led to more neutral sentiment ( $$\tau $$ = 0.39, p = 0.0027) and reduced positive sentiment ( $$\tau $$ = − 0.36, p = 0.0052). These findings highlight how robot errors and user behaviors influence HRI, emphasizing the need for adaptive robots that respond to variations in politeness and sentiment, ultimately enhancing collaboration and user satisfaction.