Physical extracurricular activities in educational child-robot interaction
PPhysical extracurricular activities in educationalchild-robot interaction
Daniel Davison and Louisa Schindler and Dennis Reidsma Abstract.
In an exploratory study on educational child-robot in-teraction we investigate the effect of alternating a learning activitywith an additional shared activity. Our aim is to enhance and enrichthe relationship between child and robot by introducing “physical ex-tracurricular activities”. This enriched relationship might ultimatelyinfluence the way the child and robot interact with the learning ma-terial. We use qualitative measurement techniques to evaluate the ef-fect of the additional activity on the child-robot relationship. We alsoexplore how these metrics can be integrated in a highly exploratorycumulative score for the relationship between child and robot. Thiscumulative score suggests a difference in the overall child-robot re-lationship between children who engage in a physical extracurricularactivity with the robot, and children who only engage in the learningactivity with the robot.
This paper discusses an exploratory study in which we investigate therelationship between a child and a robot working together to solve alearning task. In order to support children in their learning process,the relationship between the learner and the teacher or peer is crucial[15, 17]. Within this context, a child, a robot and the learning materi-als are engaged in a triadic interaction, as illustrated in figure 1. Thechild interacts with the learning materials, together with the robot, ina collaborative learning setting. Interactions between child and robotand the relationship they form can influence how the child performsin the learning task and ultimately how the child learns [5]. The tri-adic interaction consists of three distinct dyadic interactions, whichinfluence each other to greater or lesser extent: 1) interaction betweenchild and robot; 2) interaction between child and learning materials;and 3) interaction between robot and learning materials.Research in specific zones of this triadic interaction between child,robot and learning materials often focus specifically on one of thedyadic interactions, or on influences between these three dyadic in-teractions. Typical examples of research on the dyadic interactionbetween (1) child and robot are those of Kahn et al. [8] and Kanda etal. [10], who investigate children’s perceptions of the robot and rela-tionships with the robot, in an educational context. Typical examplesof studies that show how interactions between (1) child and robot influence the interactions between (2) child and learning materials are: Kory and Breazeal [13], who investigate how matching a robot’scompetence level to that of the child influences the child’s learning;and Chandra et al. [4], who show that children feel more responsiblein the learning task when working with a robot facilitator. Dept. of Electrical Engineering, Mathematics and Computer Science, HMI,University of Twente, Enschede, The Netherlands. Corresponding authore-mail: [email protected]
We investigate the dyadic interaction between (1) child and robot ,in which the child and peer-like robot engage together in a physicalextracurricular activity, in an educational context. Since the robot isnot necessarily presented to the child as a teacher, it could enrich thelearning through implicit interaction in contrast to explicit teachingand assessment. However, the study presented here does not focus onmeasuring these potential effects on learning. (cid:38)(cid:75)(cid:76)(cid:79)(cid:71) (cid:53)(cid:82)(cid:69)(cid:82)(cid:87)(cid:47)(cid:72)(cid:68)(cid:85)(cid:81)(cid:76)(cid:81)(cid:74)(cid:3)(cid:80)(cid:68)(cid:87)(cid:72)(cid:85)(cid:76)(cid:68)(cid:79)(cid:86)(cid:11)(cid:20)(cid:12) (cid:11)(cid:22)(cid:12)(cid:11)(cid:21)(cid:12)
Figure 1.
Schematic illustration of the triadic interaction between child,robot and learning materials. This triadic interaction consists of three distinctdyadic interactions: 1) child and robot; 2) child and learning materials; and 3)robot and learning materials. Each of these dyadic interactions is expected toinfluence the other two dyadic interactions to greater or lesser extent.
From Vygotsky’s theories on child development and learning, weknow that social scaffolding can be an important method for a childto transcend from his level of actual development to his level of po-tential development [17]. Previous work has shown indications thatchildren working together with a social robot interact differently withlearning materials, when compared to working with a less socialtablet [18]. We expect that a robot’s social and relational featureswill impact a child’s perception of the robot, and will influence theircollaborative long-term interactions with learning materials. For suchlong-term interactions to take place, Belpaeme et al. [1] stress the im-portance of robot adaptability to a user’s social needs. For example,Kanda et al. [9, 10] found that a friendly relationship between childand robot is one of the contributing factors for successful long-term a r X i v : . [ c s . R O ] J un se in a classroom setting.In this study we explore a possible method for enriching the child-robot relationship, through a shared extracurricular physical activity.This activity is performed in an educational context, and is intro-duced to the child as a short break from learning. The educationalassignment is based on an inquiry learning model, in which a childdiscovers properties of the learning materials using a scientific ap-proach. Hypothesis generation, experimentation, and evidence eval-uation are often described as the core processes involved in scien-tific discovery learning [12, 11, 16, 19]. A typical structured inquirylearning scenario involves multiple small assignments of increasingdifficulty, in which the learner will go through a cycle of processes.Figure 2 shows a adapted version of the inquiry cycle, which hasbeen simplified to match the skill level of primary school children.Five distinct processes are included in this structured inquiry task: 1)prepare; 2) predict; 3) experiment; 4) observe; and 5) conclude.For the specific inquiry task in this study, the learner uses a bal-ance beam to discover the “moment of force”. The balance beam actsas a weighing scale, with which the children can measure the relativeweights of various available objects. By placing the objects at differ-ent offsets from the central pivot point, they discover that the momentof force acting on the balance is influenced by both the weight of theobject and its distance from the pivot. Figure 2.
An adaptation of the inquiry cycle, simplified to match the skilllevel of primary school children.
The non-educational activity is a short physical exercise with taskssuch as: standing on one leg while raising the arms; standing on oneleg while drawing the shape of an eight in the air; and closing theeyes or spelling the alphabet. In such tasks, physical and cognitivechallenges are combined. We incorporated the cognitive challengesto make the difficulty level of the task higher (the combination alsomakes the physical activity harder to carry through). This leads toa challenging task for the child which nevertheless is very differentfrom the educational task.The robot used in this study is the Zeno R25, a humanoid robotdeveloped by RoboKind. The Zeno has five degrees of freedom in hisface, with which he can make basic facial expressions and emotionssuch as “happy”, “sad” or “surprised”. In addition, the robot is able tomove his limbs and can shift his gaze using two degrees of freedomin his neck, combined with eye movements.The goals for this study consist of two parts: Firstly, we exploreif a physical extracurricular activity affects a child’s perceived rela-tionship with the robot. Secondly, we explore novel methods for con-ducting semi-structured interviews, which can be used to measure achange in this perceived relationship. Since written questionnairesare often not suitable for children of this age, we use a mixture ofhalf-open interview questions, sociometric questions and small pic- ture tasks. In the analysis we focus not only on the given answers,but take into account the child’s reasoning behind these answers.
We focus on learners between 6 and 11 years in age. It is the purposeof this study to research whether an additional joint activity (nextto a learning activity) has positive influence on the relationship be-tween robot and child. Therefore, a randomised controlled trial hasbeen conducted between two groups of participants. The interven-tion group performed two assignments with the learning material,then they continued with performing the intermediate physical ex-ercise and subsequently they continued with two more assignmentswith the learning material. The control group completed six learn-ing task assignments together with the robot without performing theintermediate physical exercise. Consecutive learning tasks were ofincreasing difficulty, while ensuring that the difficulty of the first andlast tasks was identical in both conditions.To keep the remaining conditions as similar as possible, the over-all time each child shared with the robot was approximately similar.The physical activity is therefore approximately as time consumingas two learning assignments. In addition, the robot’s behaviour andmovements are similar in both conditions. For instance, in both con-ditions the robot uses gaze, moves his head accordingly, and movesthe body as natural as possible. Furthermore, the voice of the robotis equal in tone and emphasis, and the robot uses inviting forms ofphrasing the sentences. For instance, the robot uses the phrase “let’sdo (...)” to trigger the next activity, and uses only positive phraseswhile supporting the child, such as “well done” or “good job”.
The learning activities are based on an inquiry learning cycle, duringwhich the children go through several processes related to scientificdiscovery. They generally go through the following five processesfor each learning assignment: 1) prepare the experiment; 2) predictthe outcome (hypothesise); 3) perform the experiment; 4) observethe outcome; and 5) draw a conclusion. The adapted inquiry cycleis illustrated in figure 2. Consecutive assignments are generally ofincreasing difficulty, but all follow the same processes outlined bythe inquiry cycle.The specific inquiry activity used in this study is related to discov-ering the “moment of force”: the children use a balance beam witha central pivot, to explore the effects of weight distribution on theforces acting on the balance. For instance, they discover that plac-ing a heavy weight close to the pivot, will result in an equal force asplacing a light weight far from the pivot.
In addition to the learning activities, the children in the interventiongroup participate in a physical extracurricular activity with the robot.This physical activity was introduced by the robot as a break fromstudying. The robot invites the child to do a physical exercise to-gether, such as “Stand on one leg, and wave your arms, while recitingthe alphabet”, or “Close your eyes while moving your arms forwardand standing on one leg”. The robot performs most of the demandedmovements as well, however there are some restrictions. The robotcan move the arms up and down, close the eyes, move the head, andstand on one leg. Some of the more complex movements are impos-sible for the robot, such as “Stand on one leg and draw a ‘figure 8’ inhe air with the raised foot”. Therefore, the robot not only performsthe actions, but also verbally explains what the child should do.
The goal of this study is to evaluate the relationship between childand robot. There are several methods that can be used in order tolearn about the connection between the robot and the child. Most ofthese measures can be seen as tools with which to conduct a semi-structured interview with children. Examples of such tools can befound in the Fun Toolkit, constructed by Read et al. [7, 14]. Severalof the methods discussed below are inspired by the Fun Toolkit, andhave been adapted to match the target group’s age.The measurements can be subdivided in three main categories, allof which are collected in the form of a semi-structured interview:1. Pictorial task2. Social distance task3. Sociometric questionsFirstly, an assignment is used in which children should describetheir thoughts by means of a pictorial task. This approach supportschildren to describe situations or imperceptible concepts like rela-tionships more precisely [6]. First, the child draws himself or herselfin order to more strongly identify with the picture. Then, a picture ofthe robot is shown to the child and placed next to the child’s drawing.A collection of pictures is then shown to the child in random order,from which the child chooses the most appropriate picture to placein between the robot and the drawing of themselves. Note that theinterpretation of how to select the most appropriate picture is left tothe child. The researchers do not present the pictures in a predefinedorder, as not to bias their choice.Similar to the Smileyometer described in the Fun Toolkit [7], thepicture collection consists of smiley faces displaying certain emo-tions, as shown in figure 3. These pictures of emotions help the chil-dren describe their own emotions and feelings during the experimentand towards the robot. After choosing the most suitable picture, weask the child to explain what emotion they see on the picture, andwhy this emotion fits between them and the robot. Subsequently,the child is asked to select and explain a second picture, to allow usto explore possible nuances in the earlier answer: does the child in-cline more towards positive or negative emotions, or perhaps selectsa combination of a positive and negative emotion?
Figure 3.
Collection of emotions shown to the child to identify the child’semotions towards the robot. The pictures are presented to the child as an un-ordered collection. The child is consecutively asked to choose and explaintwo pictures, which in his or her opinion best match the interaction with therobot.
Secondly, we attempt to measure the social distance between thechild and the robot. A recognisable setting of a circle of chairs in aclassroom is shown to the children, as illustrated in figure 4. In addi-tion, the children received cartoon pictograms of the face of the robotand faces of various children. The child in the experiment choosesone of the available faces to be his or her own, while the other facesrepresent the classmates of the child. The child is then asked to place themselves, the robot and the other classmates into the room. Sincekids generally like to sit next to their friends and socially close con-tacts, we assume this setup will give an indication on how sociallyclose the child feels towards the robot. In addition, we ask the chil-dren to explain why they seat themselves and the robot in the specificlocations.
Figure 4.
Classroom with circle of chairs, wherein the children are asked toplace themselves, the robot and their other classmates.
Finally, we use questions adopted from the field of sociometry[3], supplemented with questions inspired by the work of Beran etal. [2]. More specifically, we ask children whether they would in-vite the robot to their home, whether they would tell the robot a se-cret, whether they would share food with the robot, whether the robotcould hear or see them, and whether they could be friends with therobot.The above methods are combined in the form of a semi-structuredinterview, to gain further insight in the child’s reasoning andthoughts. The half-open questions focus on the child’s opinion aboutthe task, and their perceptions of the robot and themselves in the task.From these questions a matrix is drawn to describe the relation andengagement the child has with the robot and the task. For instance, achild may answer that he or she will tell the robot a secret, anotherchild would say they don’t have a secret, but would then invite Zenoto their home. Since all these questions refer to the surrounding of thechild, it might be that one single answer is unsuitable to determinethe nature of the relationship, while a combination of questions mightlead to a more general insight on the relationship between robot andchild. Additionally, we perform a qualitative analysis of the child’sexplanations, to gain more insight in the reasoning behind selectingcertain answers.Based on the collection of these three measurements, we computean exploratory cumulative score. This score covers all the social re-lationship features that we have measured, and is composed of a sumof weighted scores for each individual measurement. We attempt tointerpret this as a summary of each individual child’s social rela-tionship with the robot, without going into detail for each individualfactor.
The overall setup of the experiment was the following: First, the re-searcher introduced the robot to the child. It was explained who theobot is and what would happen during the experiment. Then thelearning task was explained to the child. Both groups, the interven-tion and the control group, got the same introduction to the task.After the introduction, the robot guided the child through the learn-ing assignments. The assignments were additionally displayed on atablet, to supply multiple channels of communication. The consecu-tive assignments in the inquiry task were of increasing difficulty, allchildren started with the same easy assignment and ended with thesame difficult assignment. The first two assignments were the samein both conditions. The control group then received two additionalassignments of intermediate difficulty, while the intervention groupperformed the physical extracurricular activity. The participants inboth conditions then finished with the same two assignments. Thesemi-structured interview was run after the last assignment.
The experiments were conducted at a local daycare centre. Priorto participation, each child’s parent or legal guardian was informedof the activities, study goals, and data collection methods, and wasasked to fill out an informed consent form. After completing the ex-periments all children were given a central debriefing, where theycould ask questions and say goodbye to the robot. This study was ap-proved by the university’s Electrical Engineering, Mathematics andComputer Science (EEMCS) ethics board.Over the course of a week, a total of 23 children participated atthe daycare centre. In the intervention condition the 12 participants(7 boys, 5 girls) had an average age of 8 years (SD = 2.73). In thecontrol condition the 11 participants (9 boys, 2 girls) had an averageage of 8.6 (SD = 1.65).
Based on our observations, generally the children’s reactions on theexperiment were very positive and enthusiastic. Overall, they en-joyed participating in the experiment and playing with the robot.All children of the intervention group followed the robot’s sugges-tion to stand next to him to perform the physical exercise. All of themfollowed the movements and suggestions of the robot. Even difficulttasks which the robot was unable to do himself were performed bythe children. Only two of the tasks were misunderstood by a smallamount of children. Some children misinterpreted the task drawingan eight in the air with your foot. Instead of drawing an eight withtheir foot, they did it with their hand. Furthermore, we found thatchildren did not continue to spell the alphabet after the first few let-ters. The children interpreted all other parts of the learning assign-ments and the intermediate physical exercise correctly.The control group did like the task and the robot in general, butchildren indicated several times that the learning task was repetitive.None of the children of the intervention group said this about thelearning tasks in combination with the physical exercise.
Most of the children participated well in the interview and tried toanswer the questions. Some were shy, which resulted in a less de-tailed interview with unanswered questions or missing explanations.The interview started with an open question about what they likedand disliked about the experiment. All children enjoyed the learningtasks, and some mentioned it was funny to do the experiments.
What did you think about the experiment and the robot?
Allthe children answered that the assignment as well as the robot wasnice. A few children answered more specifically that they liked theprediction step of the inquiry cycle, or that they liked one of thechallenging learning tasks. At this point some children of the controlgroup indicated that the learning task was always the same. Childrenof the intervention group often said, that they liked the physical ex-ercise and overall the variety of assignments they did with the robot.
Pictorial emotion task
Children were free to interpret the emo-tions, since the collection of pictures was presented in random order.Therefore, they were first asked which emotion they saw on the pic-ture and why they chose this emotion. From these answers the child’sinterpreted emotions become apparent, as shown in table 1. Gener-ally, the interpretations were very similar. However, a small minorityinterpreted the open-mouthed smiley as surprised or amazed.
Table 1.
Children’s interpretations of their picked emotions.
Table 2.
Overview of children’s first and second picked emotions. The com-binations presented here show the amount of children who picked this specificcombination of emotions for their first and second choice. Condition 1 is theintervention group, condition 2 is the control group.
As described previously, the child would choose and explain twodifferent pictures in succession, indicating which combination of twoemotions best describe how they feel about their interaction with therobot. Table 2 shows that the children generally picked a combinationf two happy emoticons to describe their emotions. In the interven-tion group 50% of the children chose the “happiest” smiley as theirfirst choice. For the control group, 27.3% of the children chose the“happiest” smiley as a first choice. The two happy smileys have beenused 83% in the intervention group. The control group used thesetwo smileys 81%. Less happy smileys were selected 5 out of 24 forthe intervention group and 6 times out of 22 for the control group.The second combination in table 2 has been used the most in theintervention group. This combination was selected by 41.6% of theintervention group compared to 27.3% of the control group. For thecontrol group, combinations 1, 2 and 4 were each chosen in 27.3%of the cases.
Social distance
Most of the children put the robot directly nextto themselves in the class circle. When asked why, they stated thatthey liked the robot or thought it was nice to sit next to him. Someprojected a real relationship and indicated they would laugh with therobot or the robot could help them if he would sit next to them. A fewchildren placed the robot somewhere else, stating that they wantedonly their closest friends to sit next to them. In the intervention group,83% placed the robot directly next to themselves, compared to 90%of the control group.
Friendship with robot
The large majority of the children in bothgroups indicated that they could imagine a friendship with the robot:83.3% of the intervention group and 81.8% of the control group con-sidered a friendship with the robot. They reasoned with phrases like:“Zeno can do everything”, “He is very lovely”, “We look like eachother” or “We had fun together”. A low percentage of the childrendoubted before they answered and first asked, for instance, whetherthe robot could play soccer. From the intervention group 8.3% andfrom the control group 9.1% would not consider a friendship withthe robot. In summary, children of the intervention group and con-trol group showed a similar preference for becoming friends with therobot.
Age of the robot
For the age of the robot, the children reasonedvery differently. Some children observed the appearance of the robotin order to determine the age, some others included the introducedbackground of the robot, while others used their imagination. Themajority of children in both conditions stated that the robot is slightlyyounger than themselves. A few children answered that the robot iseither the same age or older. In average the age of the robot wasguessed to be 8.6 years for the intervention group. In the controlgroup the children mentioned an average age of 9.3 years.
Sharing a secret with the robot
The children were asked whetherthey would tell the robot a secret. This question measures the trust thechildren would have in the robot [3]. The children sometimes saidthey won’t tell a secret because the robot is interacting with so manychildren, there is a high chance that the robot would tell it. Otherchildren stated that they would only tell it, if the robot would not tellit to somebody else.A slight majority of both groups stated they would tell the robot asecret (58.3% in the intervention group, 54.5% the control group). Asmaller percentage would not want to tell the robot a secret (41.67%in the intervention group, 27.3% the control group). Of the controlgroup, 18.2% did not respond to the question. Most of the childrenwho would not tell the robot a secret, still said that the robot couldbe a friend.
Invitation to the child’s home
When we asked the childrenwhether they would invite the robot to their home, some of the chil-dren got very excited about this idea. Of the intervention group,91.6% wanted to invite the robot to their home, while 8.3% did notrespond to the question. From the control group, 63.6% would invitethe robot to their home, 18.2% of the children answered they wouldnot invite the robot to their home and 18.2% did not respond to thequestion.As a follow up question, we asked what the child and the robotwould do at home. The answers to this question were very broad.Of the intervention group, 75% proposed some sort of activity, all ofwhich unrelated to the experiment, while 25% did not respond to thequestion. The control group proposed 36.3% related activities and18.1% unrelated activities, while 45.5% did not respond. Clearly, thechildren that experienced the physical activity together with the robotwere able to imagine a broader range of activities they could sharewith the robot. Nonetheless the imagination of the children in generalwas very broad.
Sharing food with the robot
The children were asked whetherthey would be willing to share their food with the robot. This ques-tion was intended to trigger the child’s sociality towards the robot.All children of the intervention group wanted to share food with therobot. Of the control group, 27.3% rejected to share their food withthe robot.
Audition and sight of the robot
When asked about whether ornot the robot could see and hear, both groups were equally confusedon whether the robot has sight, audition or both. Some children in-dicated, that it was one of the researchers, who controlled the robot.Others thought the tablet or other technology enables the system tofunction autonomously.In the intervention group, 50% of the children indicated that therobot has sight and 66% stated that it could hear. Two children statedthat the robot has none of the two. In the control group, 45.5% of thechildren said that the robot can see and 54% said that it could hear.Three of the asked children stated that the robot has neither sight noraudition.
During the interview, most children commented on the robot’s ap-pearance or behaviour. Many children thought the robot was funny.They commented that the robot had a female voice and a male ap-pearance. The robot was often described as nice, smart and happy.Some children thought he was not that happy or that they expectedhim to be taller. Often, the children were surprised that he could talkand perform various complex body movements. Most of the childrenthought the robot was slightly younger than themselves, however afew kids thought he was an adult and looked older than them.The role of the robot often appeared in the interview to be very di-verse. Some children commented that they would help the robot un-derstand things, while others said the robot could help them in study-ing. One child answered that the robot could help clean his room.Another child liked that the robot was neither a teacher nor a peerstudent, but that it was nice that the robot was of “a different kind”.During the interview the children would often indicate some de-scriptive characteristics of the robot. For example, some describedthe robot the same as themselves, similar to them, or that he could doeverything. Some indicated that his movements were quite natural orhat they thought that robots could do less. On the other hand, somesaid the robot could not do all of what they themselves could do.
Due to the fact that each individual measure only looks at one veryspecific relationship aspect, we propose an additional exploratory cu-mulative analysis . For this analysis we calculate a cumulative “so-cial relationship” score for each child, based on the individual an-swers and measurements mentioned above. The expectation is thatthis cumulative score contains some information about the generalperceived social relationship they have with the robot, although wewill not be able to identify the individual aspects which influence thisrelationship.The assigned measures included most of the questions the childanswered. All these questions receive a certain score, depending onhow the child answered the question. The different scores for eachmetric can be found in table 3. Friendship gets a relatively high score,due to the fact that considering friendship is a very important criteriafor building a social relationship. Additionally, the other scores arechosen in such a way that measures with a high variance get a highervalue than the measures with a low variance. Questions such as theage are left out, as the child’s reasoning behind this has been very dif-ferent: the children’s answers did not always represent a perceptionof the robot’s age.
Table 3.
Scores applied for various measures, as used in the exploratorycumulative analysis. For each sociometric measure, a score of 0, 5, 10 or 15is added to the child’s cumulative score if answered “yes”, or 0 if the childanswered “no”. For the pictorial emotions task, a score of 1, 3, 5 or 10 isadded, depending on the selected emotion for their first and second choice.
Results from two children were excluded from this analysis, dueto their high amount of unanswered questions. The mean of the in-tervention group is 57.5 (n = 11, SD = 7.19) and the mean of the con-trol group is 45.7 (n = 10, SD = 9.59). A Wilcoxon-Mann-Whitneytest shows a significant (U = 15, p < Figure 5.
Box plot showing the distribution of the exploratory cumulativescores, for the intervention group (condition 1) and the control group (condi-tion 2).
From observations we get the impression that most of the childrenenjoyed the overall experiment and specifically the physical exercisewith the robot. However, a more detailed analysis of the recordedsessions is needed to further investigate this. So far, we base thissubjective impression on the preliminary assessment of the videosand the reactions that came from the childcare supervisors and thechildren themselves. Nonetheless, this is worth mentioning, since anenjoyable and nice task is fundamental to keeping children engagedwith learning.
We interpret the individual interview results as some of the parame-ters that describe a relationship. Although we recognise that we onlytouched upon a fraction of the factors influencing something as com-plex as social relationship building, we tried to research aspects oftrust, sociality, friendship, social distance and the children’s generalview and perception of the robot. It can be concluded that severalof the questions did not show any difference between the two con-ditions. This means that there is no clear indication that the physicalextracurricular activity influences the specific factors of the relation-ship that we asked about.A reason for the similar answers of the children in both condi-tions could be that the questions were not asked in an operative un-derstandable way. This is always a challenge with children. Theserather inoperative questions were asking about “sharing food with therobot” and the “social distance” (the chair seating assignment). Thesetwo measures were answered in the same way by almost all children.Other measures such as “an invitation to their home” or “consideringa friendship with the robot” showed more variance. However, manychildren still answered in the same way, which lead to an unclearview on these measures.Results from the interviews suggest that for some questions thereis a slight difference between the two conditions, even though thisdifference might not be very strong. In questions about “inviting therobot to the home”, “the type of playing at home”, and “sharingood”, indications can be found that the intervention group had a dif-ferent view of the robot, due to their more varied experience with therobot. This might have influenced their imaginations of what else therobot is capable of. The question whether the children would tell asecret to the robot, resulted in a mixed outcome. In terms of numbers,from the intervention group 58.3% would tell Zeno a secret. From thecontrol group, 54.5%would tell a secret. Thus, the difference is againvery small.Similar results occurred regarding the consideration of a friend-ship with the robot. The difference between the two groups wasvery small, which complicates drawing any conclusion. Therefore,it seems that most of the children would consider a friendship withZeno despite of the type of interaction they had with the robot.The emotion assignment showed some differences, however theseare difficult to interpret. The first picture they chose often describedtheir general feeling, the second picture seemed to identify whetherthey clearly tend into a certain direction in terms of emotions. Forexample, if the first picture was a normal smiling/happy emotion,then the second choice could identify the direction and verify thechoice (either very happy, or moderately happy).This was however not always the case. Sometimes, the child ex-pressed two different emotions that he or she could relate to the robot.For instance one child also chose a sad emoticon, stating that hecould also come to talk to the robot when he was feeling sad. A fewchildren interpreted the picture with the broad smile as being sur-prised. The slightly smiling emoticon has often been interpreted as a“little bit happy”, but also sometimes as “doubting”.However, it can be seen that there are more “happy” emotions se-lected by the intervention group. This shows that there is a possibilitythat, overall, the intervention group indeed felt happier about their in-teraction with the robot. The combination showing the two happiestsmileys, while choosing the happiest smiley as a first choice, waschosen 41.6% for the intervention group, compared to 27.3% for thecontrol group.The assignment where the children put themselves, their friendsand the robot into the classroom, did not result in an indication ofsocial distance. The robot was placed next to the child in almost allcases, revealing no clear difference between the conditions.Although there are some non-significant indications in favour ofthe intervention condition for some of the measures, this is not re-flected in all measurements. Whether there has been an actual differ-ence in the relationship is difficult to determine, due to the fact thatsome measures showed unclear results and in general it is difficult tomeasure relationships, especially for such a short interaction. Hence,a more robust per case interpretation is needed.
The exploratory cumulative analysis shows a difference between thetwo conditions. This analysis should be interpreted with care, though,since the weights for each measure were determined intuitively. Thisanalysis gives an estimated indication of the social relationship be-tween the child and the robot, since all respective relationship mea-sures are summarised and taken into account per child.We have shown that the scores for the intervention group are sig-nificantly higher than for the control group. However, due to the ex-ploratory nature of this cumulative score, we are unable to specifyexactly which factors have resulted in the difference found betweenthe two conditions.
Partly due to the exploratry nature of this study, we discuss severallimitations that should be taken into account when interpreting theresults.Since the child engages in an activity with the robot that is both physical and extracurricular in nature, any measured effects could beattributed to one of the following: 1) the physical nature of the activ-ity; 2) the extracurricular nature of the activity; or 3) a combinationof both. Repeating the experiment with an additional condition con-taining a “passive” extracurricular activity would allow us to explorethis effect in more detail.Some children reported the control condition as being “boring”.Structured inquiry tasks are repetitive by nature, and although theconsecutive tasks were selected in such a way that there was a steadyincrease in difficulty, this repetitive effect seemed more pronouncedin the control condition. It is therefore unclear if the children’s an-swers are a reflection of their level of enjoyment, or their perceivedrelationship with the robot. A repeated experiment could investigatethis in more detail, where all children engage in identical learningtasks and extracurricular activities. In two conditions, they would ei-ther to the extra activity alone or in collaboration with a social robot.The methods used in the semi-structured interview often showvery similar results between conditions, making it difficult to inter-pret the effects of the manipulation. Only in the cumulative analysisdo we see a difference emerging between the control and interven-tion group. More research is needed to validate the methods and de-termine underlying constructs, which will influence the weights ofthe cumulative function.
In the study presented here, a child and a robot work together on astructured inquiry learning task. We investigate whether the child’sperceived relationship with the robot is influenced by engaging ina shared physical extracurricular activity with the robot. Measure-ments are gathered using a semi-structured interview, which is com-posed of a pictorial task, a social distance task, and several sociomet-ric questions. Generally, the children seemed to enjoy working withthe robot, indicating that they would invite him home, or that theycould become friends. Results for most individual measurements areinconclusive, however. The pictorial taks, where children pick emo-tion cards that fit their relationship with the robot, seemed to givepromising results: most children gave similar descriptions of the de-picted emotions, and generally picked more positive emotions in thecondition where they engaged in the shared activity with the robot.Finally, the cumulative score that aggregates all used measures intoa single value revealed a difference between conditions, although thelimitations make it difficult to further interpret this result.Since the first insights from this study seem promising, futurework will focus on further exploration and verification of the cu-mulative analysis method proposed in this paper, as well as the indi-vidual measurement methods used during the semi-structured inter-view. Additionally, we aim to investigate how a change in relation-ship between child and robot impacts the child’s interactions withthe learning materials, and consequently the child’s learning meth-ods and learning performance.
ACKNOWLEDGEMENTS
This project has received funding from the European Union SeventhFramework Programme (FP7-ICT-2013-10) as part of EASEL underrant agreement no 611971.
REFERENCES [1] Tony Belpaeme, Paul E Baxter, Robin Read, Rachel Wood, HeribertoCuay´ahuitl, Bernd Kiefer, Stefania Racioppa, Ivana Kruijff-Korbayov´a,Georgios Athanasopoulos, Valentin Enescu, Rosemarijn Looije, MarkNeerincx, Yiannis Demiris, Raquel Ros-Espinoza, Aryel Beck, LolaCa˜namero, Antione Hiolle, Matthew Lewis, Ilaria Baroni, Marco Nalin,Piero Cosi, Giulio Paci, Fabio Tesser, Giacomo Sommavilla, andRemi Humbert, ‘Multimodal Child-Robot Interaction: Building SocialBonds’,
Journal of Human-Robot Interaction , (2), 33–53, (jan 2013).[2] Tanya N. Beran, Alejandro Ramirez-Serrano, Roman Kuzyk, MeghannFior, and Sarah Nugent, ‘Understanding how children understandrobots: Perceived animism in childrobot interaction’, InternationalJournal of Human-Computer Studies , (7-8), 539–550, (jul 2011).[3] R Busse, ‘Sociometric Assessment’, in Encyclopedia of school psychol-ogy , ed., S Lee, 520–521, SAGE Publications, Inc., Thousand Oaks,CA, (2005).[4] Shruti Chandra, Patricia Alves-Oliveira, Severin Lemaignan, Pedro Se-queira, Ana Paiva, and Pierre Dillenbourg, ‘Can a child feel responsiblefor another in the presence of a robot in a collaborative learning activ-ity?’, in , pp. 167–172. IEEE, (aug2015).[5] Vicky Charisi, Daniel Davison, Frances Wijnen, Jan Van Der Meij,Dennis Reidsma, Tony Prescott, Wouter Van Joolingen, and VanessaEvers, ‘Towards a child-robot symbiotic co-development : A theoret-ical approach’, in
New Frontiers in Human-Robot Interaction . AISB,(2015).[6] S Harter, R Pike, and Child Development, ‘The pictorial scale of per-ceived competence and social acceptance for young children’,
Childdevelopment , (6), 1969–1982, (1984).[7] Stuart Macfarlane Janet Read, ‘Endurability, Engagement and Expecta-tions: Measuring Children’s Fun’, in Interaction Design and Children ,pp. 1—-23. Shaker Publishing, (2002).[8] Peter H Kahn, Takayuki Kanda, Hiroshi Ishiguro, Nathan G. Freier,Rachel L. Severson, Brian T. Gill, Jolina H. Ruckert, Solace Shen, and Peter H. Kahn Jr., ‘”Robovie, you’ll have to go into the closet now”:children’s social and moral relationships with a humanoid robot.’,
De-velopmental psychology , (2), 303–14, (mar 2012).[9] T. Kanda, R. Sato, N. Saiwaki, and H. Ishiguro, ‘A Two-Month FieldTrial in an Elementary School for Long-Term HumanRobot Interac-tion’, IEEE Transactions on Robotics , (5), 962–971, (oct 2007).[10] Takayuki Kanda, Michihiro Shimada, and Satoshi Koizumi, ‘Childrenlearning with a social robot’, in Proceedings of the seventh annualACM/IEEE international conference on Human-Robot Interaction -HRI ’12 , p. 351, New York, New York, USA, (mar 2012). ACM Press.[11] David Klahr,
Exploring Science: The Cognition and Development ofDiscovery Processes , The MIT Press, Cambridge, 2000.[12] David Klahr and Kevin Dunbar, ‘Dual Space Search During ScientificReasoning’,
Cognitive Science , (1), 1–48, (jan 1988).[13] Jacqueline Kory and Cynthia Breazeal, ‘Storytelling with robots:Learning companions for preschool children’s language development’,in The 23rd IEEE International Symposium on Robot and Human In-teractive Communication , pp. 643–648. IEEE, (aug 2014).[14] Janet C. Read and Stuart MacFarlane, ‘Using the fun toolkit and othersurvey methods to gather opinions in child computer interaction’, in
Proceeding of the 2006 conference on Interaction design and children- IDC ’06 , p. 81, New York, New York, USA, (jun 2006). ACM Press.[15] Barbara Rogoff, ‘Cognition as a collaborative process’, in
Handbook ofchild psychology , chapter Cognition,, 679–744, John Wiley & Sons Inc,Hoboken, NJ, US, (1998).[16] Wouter R. van Joolingen and Ton de Jong, ‘An extended dual searchspace model of scientific discovery learning’,
Instructional Science , (5), 307–346, (1997).[17] L.S. Vygotsky, Mind in Society: The Development of Higher Psycho-logical Processes , Harvard University Press, Cambridge, MA, 1978.[18] Frances Wijnen, Vicky Charisi, D.P. Daniel Davison, J. van der Meij,Dennis Reidsma, and Vanessa Evers, ‘Inquiry learning with a socialrobot: can you explain that to me?’, in
New Friends 2015: the 1stinternational conference on social robotics in therapy and education .Windesheim Flevoland, (oct 2015).[19] Corinne Zimmerman, ‘The Development of Scientific ReasoningSkills’,
Developmental Review ,20