Robots at Home: Understanding Long-Term Human-Robot InteractionCory D. Kidd and Cynthia BreazealMIT Media Lab Personal Robots Group{coryk,cynthiab}@media.mit.eduAbstract— Human-robot interaction (HRI) is now wellenough understood to allow us to build useful systems thatcan function outside of the laboratory. We are studying long-term interaction in natural user environments and describe theimplementation of a robot designed to help individuals effectbehavior change while dieting. Our robotic weight loss coach iscompared to a standalone computer and to a traditional paperlog in a controlled study. We describe the software model usedto create a successful long-term HRI in this application. Wealso summarize the experimental design, analysis, and resultsof our study, the first where a sociable robot interacts with auser to achieve behavior change. Results of our study show thatparticipants track their calorie consumption and exercise fornearly twice as long when using the robot than with the othermethods and develop a closer relationship with the robot. Bothof these are indicators of longer-term success at weight loss andmaintenance and show the effectiveness of sociable robots forlong-term HRI.I. INTRODUCTIONHumans have been interacting with robots and otherautomata for many years. In the past decade, the methodical,scientific study of this interplay between man and machinehas matured into the field of human-robot interaction. Muchof the work thus far has looked at aspects of development andlearning (e.g. Breazeal’s Ph.D. thesis and subsequent work[1]); human perceptions of various portions of a robot’s ap-pearance, personality, and behaviors (work in Dautenhahn’slab [2] or Carnegie Mellon University’s HCII group [3],for example); or short-term interactions in laboratory-basedsettings (such as previous work we carried out using a varietyof robots [4], [5], [6], [7] and others at the University ofWashington on children interacting with robots[8] and atHertfordshire on how comfortable people may be near arobot [9]).The vision of the field of HRI, however, has been to createand study robots that exist in our everyday lives. The objec-tive shared by many is to build robots that will assist us inanything from the mundane tasks of cooking and cleaning tomore intellectual and social endeavors of entertainment andcaregiving. The enormous challenges presented in surmount-ing the scientific, engineering, and interaction difficulties haskept the field from creating systems capable of autonomous,sustained interaction in the real world, leaving us to buildsystems and study the resulting interactions in the microcosmof the laboratory. We have completed the first long-termstudy where a sociable robot interacts with a person overtime to effect long-term behavior change.C. Kidd is now at Intuitive Automata Inc.A. Overweight and obesityIncreasing rates of overweight and obesity in recent yearshas brought us to the point where nearly two-thirds ofour adult population falls into one of these categories. TheNational Health and Nutrition Survey in 2002 shows 65% ofthe adult population in these categories, with 30% obese and35% overweight [10], with similar trends seen in childrenand adolescents [11]. (Overweight is defined as a body massindex (BMI) of 25 up to 30 kg/m2and obese as greater than30 kg/m2.)This excess weight leads to a significant increase in manycomorbid conditions including type 2 diabetes mellitus, heartdisease, high blood pressure, and some cancers [12]. It isnoted that a reduction of even 5% to 10% of initial bodyweight can lead to a significant reduction in risk to theseconcomitant conditions [13].There is a long history of treatments trying to effect weightloss in patients. For much of this history, overweight orobese patients were given instructions or short-term treatmentand expected to lose weight and then maintain that lossindependently. Only in recent years has the medical com-munity developed an understanding of obesity as a chroniccondition that must be managed on an ongoing basis [14].Current practice uses treatments including behavioral therapy[15], lifestyle modification [16], pharmacotherapy [12], andsurgical interventions [17], as well as combinations of thesemethods [18].These improvements in health and reduction in risk isnegated when an individual regains lost weight. Unfor-tunately, nearly everyone who loses weight using currenttreatments gradually regains at least all of the weight thatwas lost during the subsequent months and years [19]. Nearlyevery study following up on weight loss shows the gradualregain of weight after the cessation of an intervention. Anexception is the group of people who are a part of theNational Weight Control Registry, a database of over 5,000people who have lost at least 30 pounds and kept it offfor over a year [20]. While the methods of weight loss andmaintenance vary across this group, some common factorsamong those successful at maintaining their weight loss is amodification of their food intake and daily exercise [21].B. Why use sociable robots for weight loss?The series of experiments we have conducted over the lastsix years has led to the desire to explore long-term human-robot interaction as well as helped to develop the reasoningas to why a robot has shown to be a more effective interactionpartner in certain settings. Taken together, they showed thepower of a robot in conducting an effective interaction andthe stronger responses that the set of robots we have usedelicited from many study participants.We have written about the application of sociable robotsto real-world problems in recent years as we have begunexploring the design and construction of such systems. Anearlier discussion on the important factors in creating arelationship [22] noted three factors that are most important:engagement of the user, trust of the system, and motivationto use the system. An early design of the system that hasbeen built is presented in a conference paper [23] that showsmany of the theoretical design decisions that underly therelationship model in this system.In this paper, we focus on the design and implementationof the software system used to create and maintain a rela-tionship for long-term HRI. For a complete discussion of thehardware and software system design and implementation;the study design, protocol, and analysis; and evaluation ofthe overall system, see Kidd’s Ph.D. thesis [24].II. AUTOM: A WEIGHT LOSS COACHA. Sociable