Natural Person-Following Behavior for Social RobotsRachel [email protected] [email protected] [email protected] uCarnegie Mellon University5000 Forbes AvenuePittsburgh, PA 15213ABSTRACTWe are developing robots with socially appropriate spatialskills not only to travel around or near people, but alsoto accompany people side-by-side. As a step toward thisgoal, we are investigating the social perceptions of a robot’smovement as it follows behind a person. This paper dis-cusses our laser-based person-tracking method and two dif-ferent approaches to person-following: direction-followingand path-following. While both algorithms have s imilarcharacteristics in terms of tracking performance and fol-lowing distances, participants in a pilot study rated thedirection-following behavior as significantly more human-like and natural than the path-following behavior. We arguethat the path-following method may still be more appropri-ate in some situations, and we propose that the ideal person-following behavior may be a hybrid approach, with the robotautomatically selecting which method to use.Categories and Subject DescriptorsI.2.9 [Artificial Intelligence]: Robotics; J.4 [Social andBehavioral Sciences]: psychologyGeneral TermsDesign, Human Factors, Experimentation, PerformanceKeywordsHuman–robot interaction, so cial robots, person tracking,p erson following1. INTRODUCTIONRob ots that operate around people in the real world needto move in coherent, easily-understood ways, so that theywill not startle or harm the people around them. In par-ticular, for robots that operate in hospitals or in nursinghomes [5, 11], we argue that such robots need to move inways that are socially acceptable to people—particularly toPermission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for profit or commercial advantage and that copiesbear this notice and the full citation on the first page. To copy otherwise, torepublish, to post on servers or to redistribute to lists, requires prior specificpermission and/or a fee.HRI’07,March8-11,2007,Arlington,Virginia,USA.Copyright 2007 ACM 978-1-59593-617-2/07/0003 ...$5.00.Figure 1: The robot, Grace, following a person downa hallway.p eople who are not roboticists. Moving in easily understoodand predictable ways will both improve people’s trust in andcomfort with the robot as well as help insure the safety ofp eople moving near the robot.Our over all goal is to develop robots with spatial socialskills—robots that navigate in ways that people understandand expect. We are interested in developing robots thatcan operate as social assistants, such as encouraging nurs-ing home residents to participate in social activities and ac-companying them to common areas. Such a robot cannotsimply drive toward a goal and expect the person to follow it;rather, it needs to engage the person and accompany himas a human would, traveling side-by-side with the person.Many social conventions exist for side-by-side accompani-ment, and a robot must be able to obey these conventions.When two people walk next to each other, they may usecues such as utterances (“let’s turn here”), gestures, andgross body movement into and away from each other’s per-sonal space. A robot also needs to offer and respond to theseso cial cues, both verbal and non-verbal. In addition, whentraveling next to a person, a robot must behave appropri-ately at bottlenecks, such as doorways or crowded hallways,when the robot may need to move ahead of, or follow be-17hind, the person. Furthermore, the robot’s behavior mayneed to be significantly different depending on the socialsituation, as side-by-side travel may take the form of purelyso cial travel or of escorting with the robot either leading orfollowing the person.As a first step toward developing robots that can accom-pany people in socially acceptable ways, we are investigatingso cial perceptions of a robot’s movement as it follows behinda person (Figure 1), as a social assistant robot might whenpassing through doorways or navigating around ob stacles.We have designed and tested two modes of person-followingto determine w hich is more natural and socially acceptable.Participants in a pilot study agreed that the robot’s behav-ior was more human-like when the robot always drove inthe direction of the person, rather than when it followedthe person’s exact path. Thus, the “direction-following”method may be the more socially acceptable behavior forthe robot to use. However, one can imagine situations whenthe path-following behavior may be more appropriate for therob ot, such as in cluttered or unsafe environments where therob ot’s sensors may not be as reliable. We therefore proposea hybrid person-following model, and suggest that the robotmay be able to learn the appropriate situations in which touse each method. Finally, we provide a discussion of whatsteps must be taken to move from a person-following robotto a robot that is capable of socially accompanying peopleside-by-side.2. RELATED WORKPerson-tracking is currently an active area of research inrob otics, and many person-trackers are demonstrated byhaving a robot follow the person being tracked. One ofthe more common methods of tracking people is to use acamera for face or color blob detection or for contour track-ing [16]. Any such camera-based method, however, is proneto suffer from varying background colors and illuminationconditions as a robot moves through various environments.Montemerlo and colleagues developed a laser-based person-tracker for mobile r obots [12]; however, their tracker requiresthe robot to have an a priori map of the environment, andwill thus not work in a new or highly dynamic environment.Kluge et al demonstrated a method of object-tracking witha laser range-finder using network optimization techniques,but this method fails if any occlusions occur [9]. Severalresearchers have investigated combining camera-based andlaser-based tracking metho ds, typically using a laser to findlegs and a camera either to detect faces [8] or to track othervisual regions of interest [10]. Using face detection, how-ever, requires that the person always face the robot, and isthus awkward for the robot to follow behind or even walknext to a person. A final method of person-tracking requiresattaching a tracking device to the person [2]. However,