Computer interfaces represented as embodied agents, either virtually as animated characters or physically as humanlike robots, utilize a powerful metaphor of everyday social interaction in order to communicate effectively with human users. One of the most promising features of embodied agents is their ability to embody humanlike attributes and make use of nonverbal conversational cues just as people do. Gaze is a particularly important nonverbal signal in social interactions and is utilized in several rich communication mechanisms with which people are intuitively familiar.

This dissertation proposes the following thesis: humanlike gaze mechanisms can enable both virtual agents and social robots to more effectively communicate with human users in situated interaction contexts. To be fully situated, these mechanisms must be tightly linked with and responsive to the user, environment, and context in which they are deployed. Four mechanisms of social gaze are discussed. The first and most basic gaze mechanism, gaze shifts, handles the coordination of both head and eye movements to direct an agent's attention from one focal point to another. The next mechanism is gaze aversion, specifying when agents should avert their gaze away from their interlocutors and what they might accomplish by doing so. By coordinating its gaze with gaze motions tracked from a human collaborator, the agent can be become more tightly situated in the task and improve collaborative outcomes. This goal is captured in a mechanism referred to as gaze coordination. Finally, it is important to consider that a one-size-fits-all approach limits an agent's ability to account for cultural and individual differences across human users. The final mechanism presented in this dissertation, gaze adaptivity, demonstrates how the timing of an agent's gaze shifts can be manipulated in order to express extroversion or introversion, and how this personality expressed via gaze can be matched to a user's personality in order to improve motivation in a rehabilitation setting.

This dissertation makes a number of design, systems, and empirical contributions to research on human-robot interaction (HRI), intelligent virtual agents (IVA), human-computer interaction (HCI), multimodal interaction, and human communication. Overall, this dissertation contributes a set of gaze models that embody humanlike gaze mechanisms situated in specific interaction contexts, systems that implement these gaze models on virtual and physical agent platforms, and a number of user studies that demonstrate the effectiveness of these models and the general importance of well-designed gaze behaviors for achieving powerful social and cognitive outcomes for human users.