Validation of a Behavioral Robotics Framework for Social Head Gaze Vasant Srinivasan Texas A & M University College Station, TX USA vasants@cse.tamu.edu Cindy L. Bethel Mississippi State University Starkville, MS USA cbethel@cse.msstate.edu Robin Murphy Texas A &M University College Station, TX USA murphy@cse.tamu.edu Clifford Nass Stanford University Stanford, CA USA nass@stanford.edu ABSTRACT This 1 x 3 between-participant study (N=93) validates the correctness and computability of a behavioral robotics frame- work for social head gaze. The framework unifies previous studies into a collection of mappings of the engagement E for one of 5 social contexts C to one of 6 head gaze acts A. The framework was implemented with manual coding and autonomous computation of gaze acts synchronized with speech for a 911 call dispatch dialog between rescue robot and victim. Both conditions were rated positively (p<.05) for comfort, likability, robot caring, robot empathy, human- like behavior, understanding of robot behavior when com- pared to the no gaze condition. Only the autonomous condi- tion showed statistically significant improvements (p<.05) over the no gaze condition for robot integrity, robot loy- alty, at ease, engagement and positive feelings. The behav- ioral framework contributes a foundation for a comprehen- sive computational model of gaze, a formal vocabulary for social gaze, six new measures, and a novel mechanism for inferring social context from sentence structure. Keywords Agents and Intelligent Systems, Human-Robot Interaction, Gaze, User Study and Evaluation 1. INTRODUCTION This paper describes a 1 x 3 study (N=93) study that val- idated the correctness and computational viability of a be- havioral framework for social head gaze in robots. A be- havioral robotics approach was chosen because it allows the phenomena to be formally expressed using conventions well- established in artificial intelligence for robotics, encourages modularity and extensibility with behaviors as building blocks, and enables a robotic implementation. The framework was populated by unifying 12 known studies on social head gaze in humans and robots into a set of engagements, E, for five social contexts, C, and six gaze acts, A. The engagements * Proceedings of the “Gaze in Human-Robot Interaction” Workshop held at the 7th ACM/IEEE International Con- ference on Human-Robot Interaction (HRI 2012). March 5, 2012, Boston, MA, USA. Copyright is held by the au- thor/owner(s). were implemented as five production rules for a robot that used text-to-speech language generation. 2. RELATED WORK No study appears to have comprehensively addressed social head gaze, either to capture all the contexts or how it is generated. However, previous work [17] has identified twelve studies Imai et al. 01, Imai et al. 02, Breazeal et al. 03, Sidner et al. 05, Mutlu et al. 06, Yamazaki et al. 08, Mitsunaga et al. 08, Mutlu et al. 09(a,b), Staudte and Crocker 09(a,b), Sidner et al. 11 [10, 4, 16, 13, 14, 19, 15, 18, 11, 9, 21, 12] that have addressed a model or some aspect of generating social head gaze for robots. Collectively, these twelve studies describe five distinct types of engagements for five social contexts that are expressed with one or more of six distinct head gaze acts that populate the behavioral framework described in the next section. The collection of engagements of humans and robots where head gaze has been shown to be used effectively, or social contexts, are: Establishment of Agency, where head gaze reinforced human- like presence and general aliveness [4]. Communication of Social Attention, where the robot showed interest by looking at the human [10, 4, 16, 13, 21, 11, 9]. Regulation of the interaction process, where head gaze sig- naled conversational participation and managed turn-taking, especially for verbal communication [13, 14, 7, 5, 9]. Manifestation of Interaction Content, where head gaze cued goal achievement such as looking at an object to confirm it was the one that should be acted upon [10, 18, 19, 16, 4]. Projection of Mental State, which for the purposes of this paper includes expressing confusion, emotions, or intent [4, 15, 7]. The studies relied on a subset of the following six distinct “head” movements or gaze acts listed below. Of these six gaze acts, the first three (fixation, aversion, and concur-