HRI researchers interested in social robots have made large investments in humanoid robots. There is still sparse evidence that peoples' responses to robots differ from their responses to computer agents, suggesting that agent studies might serve to test HRI hypotheses. To help us understand the difference between people's social interactions with an agent and a robot, we experimentally compared people's responses in a health interview with (a) a computer agent projected either on a computer monitor or life-size on a screen, (b) a remote robot projected life-size on a screen, or (c) a collocated robot in the same room. We found a few behavioral and large attitude differences across these conditions. Participants forgot more and disclosed least with the collocated robot, next with the projected remote robot, and then with the agent. They spent more time with the collocated robot and their attitudes were most positive toward that robot. We discuss tradeoffs for HRI research of using collocated robots, remote robots, and computer agents as proxies of robots.

When a robot provides direction--as a guide, an assistant, or as an instructor--the robot may have to interact with people of different backgrounds and skill sets. Different people require informat on adapted to their level of understanding. In this paper, we explore the use of two simple forms of awareness that a robot might use to infer that a person needs further verbal elaboration during a tool select on task. First, the robot could use an eye tracker for inferring whether the person is looking at the robot and thus in need of further elaboration. Second, the robot could monitor delays in the individual's task progress, indicating that he or she could use further elaboration. We investigated the effects of these two types of awareness on performance time, selection mistakes, and the number of questions people asked the robot. We did not observe any obvious benefits of our gaze awareness manipulation. Awareness of task delays did reduce the number of questions participants' asked compared to our control condition but did not significantly reduce the number of select on mistakes. The mixed results of our investigation suggest that more research is necessary before we can understand how awareness of gaze and awareness of task delay can be successfully implemented in human-robot dialogue.

Humanoid robots offer many physical design choices such as voice frequency and head dimensions. We used hierarchical statistical mediation analysis to trace differences in people's mental model of robots from these choices. In an experiment, a humanoid robot gave participants online advice about their health. We used mediation analysis to identify the causal path from the robot's voice and head dimensions to the participants' mental model, and to their willingness to follow the robot's advice. The male robot voice predicted impressions of a knowledgeable robot, whose advice participants said they would follow. Increasing the voice's fundamental frequency reduced this effect. The robot's short chin length (but not its forehead dimensions) predicted impressions of a sociable robot, which also predicted intentions to take the robot's advice. We discuss the use of this approach for designing robots for different roles, when people's mental model of the robot matters.

Human-robot interaction could be improved by designing robots that engage in adaptive dialogue with users. An adaptive robot could estimate the information needs of individuals and change its dialogue to suit these needs. We test the value of adaptive robot dialogue by experimentally comparing the effects of adaptation versus no adaptation on information exchange and social relations. In Experiment 1, a robot chef adapted to novices by providing detailed explanations of cooking tools; doing so improved information exchange for novice participants but did not influence experts. Experiment 2 added incentives for speed and accuracy and replicated the results from Experiment 1 with respect to information exchange. When the robot's dialogue was adapted for expert knowledge (names of tools rather than explanations), expert participants found the robot to be more effective, more authoritative, and less patronizing. This work suggests adaptation in human-robot interaction has consequences for both task performance and social cohesion. It also suggests that people may be more sensitive to social relations with robots when under task or time pressure.