User Acceptance of Autonomously Generated Social Head Gaze

This paper shows that autonomous generation of head gaze using our novel behavioral framework maintains a high level of user acceptance equivalent to that from manual coding of head gaze for robotic systems. The current practice for im-plementing social head gaze is to either use manual coding, which requires significant human effort and restricts real-time dialog, or autonomous generation through ad hoc im-plementations addressing only one or two of the five known social contexts. Our behavioral robotics framework approach autonomously produces head gaze acts without semantic un-derstanding or modeling of the agents ’ theory of mind using eight novel affordances from sentence structure and time in-tervals between exchanges along with coordination functions to trigger behaviors that generate head gaze acts loosely syn-chronized with speech. The correctness and sufficiency of the autonomously generated head gaze acts was validated by a 93 participant study in which participants played the role of trapped victim in a simulated disaster setting and interacted with a rescue robot for 15 minutes. The autonomous condi-tion was evaluated more positively (p < 0.05) compared to no gaze condition for user’s affective state and perceptions of the robot. There was no statistical significance between the autonomous and manual conditions for any of the mea-sures. Our method is real-time, repeatable, extensible, and reduces the work in Wizard of Oz type studies