Convenience vs. Reliability? Evaluation of Human-Robot Interaction Preferences in a Production Environment

This study examines the dynamics of human-robot interaction in an industry-related scenario, with a specific emphasis on how different error rates of a speech recognition system affect user preferences with respect to the kind of interface that is used for interaction. In modern production contexts, the growing number of collaborating robots (cobots) requires a thorough understanding of user interactions in order to build effective and user-friendly human-robot interaction approaches. This research is motivated by two important factors: (1) the continuous transformation of work due to automation, which leads to a shift in the responsibilities of human workers from manual labor to monitoring and directing, and (2) the lack of research that investigates the impact of speech interaction on trust in situations involving human-robot interaction. In consideration of this context, we conducted a user study using a closed wizard-of-oz design, in which users engage with a cobot that is fitted with a sanding end-effector in a noisy (75 dB) environment. The users could either use voice interaction or a robot teach panel to delegate tasks to a cobot while they had to complete a mental taxing task. The study evaluates the influence of different speech recognition error rates on user preferences by comparing the ease of use of voice-based interaction with the reliability of graphical user interfaces. Contrary to expectations, the findings contradict the predicted negative correlation between error rates and SUS scores. The use of the robot teach panel increases once speech recognition error rate exceeds 60%, indicating an overall preference of a convenient voice-based human-robot interaction even in noisy production facilities.