Robotics and Autonomous Systems 58 (2010) 322–332 Contents lists available at ScienceDirect Robotics and Autonomous Systems journal homepage: www.elsevier.com/locate/robot Affective social robots ✩ Rachel Kirby a,∗ , Jodi Forlizzi b , Reid Simmons a a Carnegie Mellon University, Robotics Institute, 5000 Forbes Avenue, Pittsburgh, PA 15213, USA b Carnegie Mellon University, Human–Computer Interaction Institute and Design Department, 5000 Forbes Avenue, Pittsburgh, PA 15213, USA article info Article history: Received 25 July 2007 Received in revised form 4 September 2009 Accepted 21 September 2009 Available online 30 September 2009 Keywords: Human–robot interaction Social robots Emotions Moods Affective modeling abstract For human–robot interaction to proceed in a smooth, natural manner, robots must adhere to human social norms. One such human convention is the use of expressive moods and emotions as an integral part of social interaction. Such expressions are used to convey messages such as ‘‘I’m happy to see you’’ or ‘‘I want to be comforted,’’ and people’s long-term relationships depend heavily on shared emotional experiences. Thus, we have developed an affective model for social robots. This generative model attempts to create natural, human-like affect and includes distinctions between immediate emotional responses, the overall mood of the robot, and long-term attitudes toward each visitor to the robot, with a focus on developing long-term human–robot relationships. This paper presents the general affect model as well as particular details of our implementation of the model on one robot, the Roboceptionist. In addition, we present findings from two studies that demonstrate the model’s potential. © 2009 Elsevier B.V. All rights reserved. 1. Introduction Social robots, such as those that operate in healthcare institutions and in museums, need to communicate with people in ways that are natural and easily understood, even by non- roboticists. We believe that one way to improve these interactions is to have robots display changing moods and emotions, just as humans do. This paper describes a generative model of affect that attempts to strongly mimic how people emote in order to produce as natural-seeming a system as possible. The model is designed particularly for robots that interact with people over long periods of time. As such, our focus is on modeling the long-term aspects of, and interactions between, emotions, moods, and attitudes. We have implemented our affective model on the Roboceptionist, a robot that interacts with people on a daily basis [1]. In addition, we have run several experiments to demonstrate the model’s use in social situations, which show that people do recognize emotional expressions on the robot’s face (Fig. 1) and that such expressions can significantly influence how people interact with the robot. 1.1. Human interaction Affect, such as mood and emotion, plays a major role in human interaction. Quite often,emotional reactions are caused ✩ This work was funded in part by an NSF Graduate Research Fellowship to the first author and by NSF grants #IIS-0329014 and #IIS-0121426. Portions of this work have appeared previously in Gockley et al. (2006) [34,37]. ∗ Corresponding author. Fax: +1 412 268 5571. E-mail addresses: rachelg@cs.cmu.edu (R. Kirby), forlizzi@cs.cmu.edu (J. Forlizzi), simmons@cs.cmu.edu (R. Simmons). by social interactions, influenced by societal and cultural norms, or used to communicate desires to other people [2]. Emotions carry conversational content, allowing conversational partners to form common ground and communicate more effectively [3]. For instance, an expression of sadness—facial, vocal, or behavioral— may indicate a desire to be comforted. Furthermore, what mood a person is in has a strong impact on how that person interacts with others [4]; for example, people who are interacting may ‘‘catch’’ each other’s moods and emotions, unconsciously matching their own emotional states to their conversational partners’ [5]. Frijda argues that the primary reason for social interaction is, in fact, to experience emotions, which serve to form a ‘‘sense of coherence with others’’ [6]. Suppression of emotions can be highly detrimental to relationship forming and is disruptive to conversations [7]. A well-studied effect of human–computer interaction is that people tend to treat computer agents in the same way that they treat other people, forming social relationships with them [8]. We believe that this tendency to form social relationships with computers will also apply to robots, perhaps even more so. If that is the case, then people will respond to a robot’s emotions as though the robot were human, and will expect the robot’s emotional responses to be consistent across multiple interactions. 1.2. Human–robot social interaction In recent years, the robotics community has seen a gradual increase in social robots, that is, robots that exist primarily to interact with people. Museum tour-guide robots [9] and robots that interact with the elderly [10] demonstrate not only the 0921-8890/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.robot.2009.09.015