Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

An Evaluation of Human Conversational Preferences in Social Human-Robot Interaction

An Evaluation of Human Conversational Preferences in Social Human-Robot Interaction Hindawi Applied Bionics and Biomechanics Volume 2021, Article ID 3648479, 13 pages https://doi.org/10.1155/2021/3648479 Research Article An Evaluation of Human Conversational Preferences in Social Human-Robot Interaction Chapa Sirithunge , A. G. Buddhika P. Jayasekara , and D. P. Chandima Intelligent Service Robotics Group, Department of Electrical Engineering, University of Moratuwa, Moratuwa 10400, Sri Lanka Correspondence should be addressed to Chapa Sirithunge; hpcschapa@gmail.com Received 8 January 2020; Revised 18 September 2020; Accepted 29 January 2021; Published 23 February 2021 Academic Editor: Francesca Cordella Copyright © 2021 Chapa Sirithunge et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. To generate context-aware behaviors in robots, robots are required to have a careful evaluation of its encounters with humans. Unwrapping emotional hints in observable cues in an encounter will improve a robot’s etiquettes in a social encounter. This article presents an extended human study conducted to examine how several factors in an encounter influence a person’s preferences upon an interaction at a particular moment. We analyzed the nature of conversation preferred by a user considering the type of conversation a robot could have with its user, having the interaction initiated by the robot itself. We took an effort to explore how such preferences differ as the factors present in the surrounding alter. A social robot equipped with the capability to initiate a conversation is deployed to conduct the study by means of a wizard-of-oz (WoZ) experiment. During this study, conversational preferences of users could vary from “no interaction at all” to a “long conversation.” We changed three factors in an encounter which can be different from each other in each circumstance: the audience or outsiders in the environment, user’s task, and the domestic area in which the interaction takes place. Conversational preferences of users within the abovementioned conditions were analyzed in a later stage, and critical observations are highlighted. Finally, implications that could be helpful in shaping future social human-robot encounters were derived from the analysis of the results. set of restrictions to abide by during an interaction with the 1. Introduction robot. Intelligence in initiating conversations at right occasions Acceptance of service robots in social environments has inspired many researchers to explore human tendencies is highly appreciated in achieving a robot’s context-aware behavior. Many users prefer to interact with robots, with when dealing with social robots. Conventional service robots speech [8], and the nature of speech must be friendlier and deployed in social environments are expected to support human-like. In other words, these robots are preferred to daily routine tasks such as cooking, cleaning, and taking care determine when to interact and when not to. During the of health [1–3], but modern assistive robots must also have cognitive skills to maintain a friendly and human-like inter- interaction, this natural behavior enhances the cohesion between robot and its nonexpert user. Such robotic systems action with the humans they daily meet [4]. For these robots, to be accepted by its user for a long duration, certain human- which can replicate complex human behavior in order to play the role of a close contact such as “friend” rather than a “ser- like qualities have to be embedded. Perceived sociability, cog- nitive skills, and adaptation are found to be the key factors vant” are being developed [9–13]. On the one hand, humans considered in long-term acceptance of a social robot [5]. Fur- prefer robots with at least some context awareness as well, in thermore, making right interaction decisions is equally addition to performing a predefined set of tasks. As a result, important in playing the role of a companion rather than robots can collaborate with people without disturbing them being just a service provider [6, 7]. These features ease when they are engaged in an activity. On the other hand, dealing with the robot without stressing out its user with a when robots have an instinct of how to co-op with the 2 Applied Bionics and Biomechanics situation, users do not need to stress themselves with a set of the conventions of conversation to connect with their human predefined behaviors that are perceivable by the robot. counterparts. In [18], the authors have investigated occasions Therefore, robotic systems, which adapt to the circumstances in which active collaboration between a robot and a human is in a certain encounter, are demanding in this era. Hence, required in service applications. Perceived connection between the human and the robot becomes effective, and social intelligence is an emerging requirement in human- the mutual interest grows when the two participants can robot interaction concerning social environments. Likeliness of the robot being accepted as a conversational understand the intentions of each other in a conversation; for instance, when to continue interaction and when to stop. partner depends on the environment as well as the current task of the user. The study in [14] ensures how the social As per this study, understanding the situation of a human is a demanding feature in robot’s acceptance within a sociable behavior of a robot improves the acceptance from its users human environment. as a companion in social human-robot domains. We investi- gate how the nature of interaction initiated by the robot Certain features and behaviors embodied in robots make an impact on people’s willingness to engage in at affects its acceptance by a human. Subsequently, a set of affect-based types of conversation were selected and imple- least a short interaction with the robot. The work explained in [19] has presented a set of social rules for robot behavior mented on a service robot in a simulated social environment (a “robotiquette”) that is comfortable and acceptable to in which few users were present or only a single user was present. This study is intended to find human tendencies humans. According to that, the conceptual space of HRI (human-robot interaction) studies expects a robot compan- towards interaction in different situations and hence will pro- vide means of engraving social skills into a robot’s behavior ion in a home environment to “do the right things” and “fulfil its tasks” in a manner that is acceptable and comfortable to before utilizing in human environments. To support this humans. Furthermore, real-time performance of the robot approach, we conducted a wizard-of-oz study to explore the nature of human conversation with the presence of a service which follows human social conventions and norms is more likely to be accepted for a long duration by humans [20]. robot when several factors in the environment vary. Factors which are more likely to have an influence upon the conver- In [21], a robot which is also an intelligent weight loss coach has been implemented. This makes an excellent exam- sational preference of humans are selected. The type of inter- ple for the situation perception embodied in the robot itself action preferred by the user was used as a mediator to perceive user situation and the level of interest towards the and hence has been exploited to reduce obesity. During this case, results show that the robot is accepted for a long-term interaction initiated by the robot. Responses observed during the study can be used to upgrade existing robots’ perception interaction by its users. However, this robot is not fully capa- of human behavior. Hence, this will allow a robot to be a ble to identify user behaviors which are not related to physi- cal health. But the fact that a higher social intelligence as well successful companion to the human without violating user expectations. as a greater acceptance from the user can be achieved by engraving abilities related to emotional and instinctive As humans establish emotional ties with whom they interact, this fact may remain the same for a human-robot behavior cannot be neglected in this scope [22]. Furthermore, interaction as well. Hence, the proactive, social means of there are many robotic systems to carry out a smooth conver- sation but the capability of these systems is limited to only engagement are expected from a robot in such a scenario [15]. One aspect in developing mechanisms to enhance social after the initiation of a conversation but not before the con- versation [23]. In [24], conversation was used as a part of intelligence in robots is to improve the user experience with such robots. A similar study was conducted to analyze con- assistance for Alzheimer patients in addition to therapy. This versational preferences of schizophrenia patients with robots study has identified robot’s knowledge of the location, patient’s history, type of disease, etc. and these parameters in [16]. Even so, only the context covers only patient-robot interaction, not human-robot interaction entirely. are important to decide the level of interaction between the patient and the robot. According to the survey by Lorenza et al. [17], the cogni- Law et al. [25] present a similar approach towards tive, emotional, and behavioral examination of human responses have an impact upon a robot’s behavior. Therefore, understanding one aspect in this regard. The authors have conducted a human study by means of a wizard-of-oz to measure this impact adequately beforehand, it is important for the robots to be cautious upon the factors which affect a experiment, to assess the level of curiosity aroused in humans when dealing with an assistive robot. Study con- human’s interaction decisions during a particular encounter. firmed the fact that the human curiosity considerably From this study, we intend to lay a justifiable basis to bring several such observable factors that can be used by a robot to changes when the intelligence of the robot is higher. More- over, it is found that the social acceptance of a service robot evaluate an encounter before robot-initiated interaction. We considered observable cues from humans as well as their sur- increases when a robot is able to perceive the very needs of a user and act accordingly [26]. Results of the human study roundings, which were likely to have an impact towards in [27] verify that humans prefer user adaptive dialogs in responses generated by a human in a certain instance. conversations even with a robot. The work explained in [28] is a promising example of the growing rapport 2. Related Work between humans and robots with such an intelligence. In order to participate in collaborations with people, robots Human further prefers the companionship build through interactive conversation between him and the robot; in must not only see and talk with people but also make use of Applied Bionics and Biomechanics 3 psychophysiological factors contribute majorly in deciding addition to the service purposes, most robots are intended for [29]. the level of interaction readiness within a human. Objects A method for attention estimation is proposed in [30]. and other humans in the surrounding, obstacles, etc. make Authors have used human pose and speeds of specific angu- the list of factors in the environment which have to be lar joints to identify the nonverbal interaction demanding of perceived by the robot. Figure 1 demonstrates this idea. a user. These two parameters were evaluated using a fuzzy logic-based mechanism to evaluate the interest level of a 3.2. Theory of Planned Behavior. Out of many psychologi- human towards a robot. However, these are not the only cal theories behind human tendencies, theory of planned parameters which define the nonverbal interaction demand- behavior lays a reasonable, yet justifiable basis for the dif- ing of a human. According to a review assessment performed ferences in human behavior under various circumstances in [31], wizard-of-oz is an effective mechanism to study in the environment. human tendencies based on nonverbal behavior. WoZ stud- According to the theory, one’s believes are linked to ies are effective in assessing such interactive sessions in a his/her behavior. This makes reasoned actions based on a short period without unnecessary preparations beforehand. restricted or controlled behavior. An individual’s intention An example scenario which evaluates speech-based interac- of a certain behavior at a specific time, a place, etc. is based tive interfaces is presented in [32]. In [33], Sidner et al. fur- on the regulations that humans follow by. These will take ther elaborate that the dialog features are important in three forms: behavioral, normative, and control. The theory accepting a robot for a long duration during human-robot of planned behavior comprises of six constructs that collec- interaction. tively present actual control of a person over the behavior The above discussed systems access a limited number of [35]. These constructs are stated briefly as follows. cues from its users and their surroundings before making interactive decisions. Still, there are several other factors to (1) Attitudes—degree to which the individual has a be considered before generating responses in a human- favorable or an unfavorable evaluation upon the robot social encounter [34]. We selected several such factors behavior of interest which are likely to have an impact upon user responses dur- ing an encounter and investigated whether these factors (2) Behavioral intention—motivational factors that actually make such an impact. Hence, the findings of the influence the behavior study can be used to improve the conceptual basis of reason- (3) Subjective norms—belief about whether behavior will ing in future social robots. be approved by peers and people of importance In our work, the importance of understanding user situ- ation is evaluated using such a WoZ approach. Factors used (4) Social norms—customary behavior in a group of to define user situation were the activity or the current task people belonging to a cultural context of the user, number of people around the user, and the type (5) Perceived power—the behavioral control over these of area of the house. Conversational preferences when these factors that may facilitate or impede performance factors change were analyzed with the help of a domestic ser- vice robot platform placed in a simulated social environment. (6) Perceived behavioral control—person’s perception of Since our work is based on verbal human behavior, a ease or the difficulty in performing the action WoZ experiment will have the capability to explore unex- pected tendencies in human behavior prior to an interaction. This concept can be related to human-robot scenario as In this work, we tried to investigate human behaviors that follows. Due to the factors in the environment, user’s percep- can be used as cues for a robot to perceive user situation prior tion of the environment or the surrounding may subject to to an interaction. It is expected that these findings will help change, depending on his/her beliefs. Hence, the reaction improve social intelligence of a social robot for the purpose towards robot’s conversations may change in different of caretaking and simultaneously providing emotional scenarios. Three factors which are most likely to affect the support through interaction. user response are considered in this study. These factors are selected from user and environment aspects. The purpose of this is to evaluate the human behavior during these situa- 3. Theoretical Approach tions. These factors are listed below. 3.1. Robot’s Perception of the Environment. A situation between a human and a robot consists of the robot itself, (i) Task of the user—e.g., having a snack, cleaning, and the user (human), and the environment (objects and space) engaged in a desk activity around the robot and the user. When the robot intends to (ii) People in the surrounding—alone or surrounded by perceive such a situation, it first has to identify interactive few people factors within itself, the environment, and the user. Factors within the robot itself include the dialog patterns the robot (iii) Type of area in the domestic/social environment—liv- generates, maintaining an interactive distance in between, ing room, bed room, or kitchen and displaying appropriate behavior, etc. Factors within the user will be numerous, but emotions, social norms, beliefs, These three factors contribute to evaluate mainly atti- personality traits, user’s activity at that moment, and other tudes, subjective norms, and perceived behavioral control 4 Applied Bionics and Biomechanics Perception criteria Example scenarios encountered in a social environment with the presence of humans and robots as perceived by the above the- 1.Self (robot) ories are given in Figures 2 and 3. In occasion 1 (Figure 2), the 2.User (human) Bed room user gave priority to an interaction with the robot, but in occasion 3.Environment 2 (Figure 3), the user gave priority to her current activity. In both User the occasions, various factors within the environment and the user itself will affect her response. Robot 4. Experiment Living room 4.1. Setting and the Research Platform. The experiment was conducted in a simulated social environment in the labora- tory. Participants were students, nonacademic staff members of the university, and some outsiders in the age range 19-58 (Mean-28.45, SD-9.02) who volunteered the study. There were 37 participants, and they were in good health condition Figure 1: An example domestic environment is shown. In this without any physical defects which will alter their reactions scenario, the user is involved in a desk activity in the bed room. In during the study. More than half of the participants did not order to understand the whole situation, the robot has to be have a technical background in education, majors, or knowledgeable on three aspects: itself, the user, and the research related to Engineering. The gender of the user was surrounding environment. Factors related to these three aspects are marked as 1,2, and 3, respectively. not included within the scope of this study. Upon arrival, the users were given instructions regarding the tasks they should complete but they were not aware of the fact that they out of the six constructs of the theory of controlled behavior. are intended to talk to the robot but they are instructed to It is assumed that the type of interaction preferred by the user respond towards the robot if the robot initiates an interac- change when these factors change. We evaluated this fact tion. They were not knowledgeable about the exact intention through the human study conducted in the form of the of the experiment because that will cause a bias response WoZ experiment. from users towards the robot. Hence, the participants were An application of the two theories: occasion 1 (a), user instructed to perform a given activity in the way they are used was working and has no idea about the presence of the robot, to perform that before. (b) notices the presence of the robot as it moves and as a The experiment was conducted using a service robot result, the user looks at the robot, and (c) stops the work called MIRob. The robot is visually and verbally capable and gives attention to the robot while it approaches the user. and has the ability to approach a user, make a conversation, An application of the two theories: occasion 2 (a), user and handle objects. This is a Pioneer 3DX MobileRobots was working and has no idea about the presence of the robot, platform equipped with a Cyton Gamma 300 manipulator (b) notices the presence of the robot as it moves and looks at and a Kinect camera for vision. The maps required to navi- the robot, (c) user averts her gaze and give attention to the gate around were created with Mapper3 Basic software. The work, and (d) engage in the work again. platform is equipped with a microphone and a speaker to listen to and respond its users. This platform is shown in 3.3. Theory of Reasoned Action. The theory explains that Figure 4. there is a relationship between one’s attitudes and actions [36]. Hence, the theory is used to predict the behavior of a 4.2. Procedure. The selected user was allowed to engage in a human in a particular scenario, based on the preexisted atti- certain task, and the robot was allowed to approach the user tudes and behavioral intentions of that individual. That indi- to initiate a conversation with him/her. The user was advised vidual’s expectations upon the outcomes of a behavior to complete a certain task and if the robot talks to him/her, to controls his/her decision to adopt that behavior. This fact is talk back. The set of tasks to be performed by the users was deployed in exploring the tendencies in human behavior in predefined. There were separate lists of tasks to be performed the presence of the robot used in the WoZ study. In such a in the living room, bed room, and kitchen. The participant or situation, there are few stages which a user goes through; the user was knowledgeable on the tasks that are to be per- for instance, noticing the presence of the robot, responding formed in each living area. The tasks were selected so that towards robot which approaches towards him/her, initiating there will be at least three tasks performed in each area. These a conversation with the robot, or responding to a conversa- tasks are the most common to that particular social or tion initiated by the robot can be stated as the usual stages domestic environment, and few tasks selected for the study of interaction in such a situation. The user’s conversational are listed in Table 1. While the user was engaged in a task, preferences might change according to his/her attitudes, the robot was remotely guided towards him/her and was beliefs, and expectations in such an instance. These attitudes, allowed to initiate an interaction in the ways given below. beliefs, and expectations may subject to change depending on This set of experiments was conducted over a period of 7 the factors present in the environment, and this fact is going days, so that the participants had enough time in between to be evaluated through this study. tasks. This prevented participants getting exhausted and Applied Bionics and Biomechanics 5 (a) (b) (c) Figure 2: An application of the two theories: occasion 1 (a), user was working and has no idea about the presence of the robot, (b) notices the presence of the robot as it moves and as a result, the user looks at the robot, and (c) stops the work and gives attention to the robot while it approaches the user. (a) (b) (c) (d) Figure 3: An application of the two theories: occasion 2 (a), user was working and has no idea about the presence of the robot, (b) notices the presence of the robot as it moves and looks at the robot, (c) user averts her gaze and give attention to the work, and (d) engage in the work again. hence generating biased, involuntary responses towards the CON-Long conversation. robot. People in the surrounding were not participants but How a conversation is categorized into these types is one or two members from the set of experimenters. And all shown in Figure 5. As the conversation extends, the type of 37 users participated in the experiments at least for three dif- conversation shifts from NI to a CON. The robot will not talk ferent tasks. As the number of participants was 37, each con- to its user during NI. In GR, the robot will only greet the per- ducted 12 tasks, 2 times (alone and with the presence of a few son and navigate away. The greeting will just be a single sen- tence saying “good morning,”“hey,”“hello,” etc. In SER, the others in the surrounding), the experiment was conducted 888 times throughout a week. We kept gaps in between robot will ask to deliver something for the user, as an assis- experiments to avoid users repeating the same response over tance to his/her current task. This will be approximately four and over by practice. The map of the environment was prede- sentences maximum in the entire conversation. In the TLK, fined in the simulation. Therefore, the robot navigated to the the robot will say a few additional sentences other than greet- target positions and its orientation which were defined by the ing and sometimes will ask if the user wants something. Such operator. In this scenario, the target position of the robot was a conversation consisted of about 5-7 sentences. All the con- a point within the interactive area near the user. For the ease versations longer than that were considered as CON. Such of future referencing, these types of conversational prefer- conversations cover a broader scope of topics as well as these ences are abbreviated as follows. existed for a longer duration. Therefore, the duration of the NI-No interaction conversation depended on the type of conversation robot GRT-Greeting had with a user. In all the occasions, the robot stopped con- SER-Asking to deliver a service tinuing the conversation depending on the curiosity of the TLK-Small talk user to engage with the robot or when the conversation seems 6 Applied Bionics and Biomechanics Conversation begins R: Good morning! H: A very good morning! Greeting R: You want me to get something for you? H: No, Thank you. Asking to deliver a service R: Okay. Are you tired? H: No, I’m just relaxing R: Okay. Small talk R: Is there something you want me to talk about? H: Hmm, There’s nothing specific, but I’d like a conversation. R: May I play a song? H: I would rather hear a poem. R: Good choice. A poem about nature? H: That’ll be great. Go on. Figure 4: Service Robot platform used in the experiment: MIRob. R: Have a nice day! H: You too! Long conversation Table 1: Some of the tasks selected for the study. Conversation ends Living area Task Figure 5: The nature and the length of conversations determine the Resting while sitting “type of conversation” existed at a certain occasion. Here “R” and Reading while sitting “H” represent the robot and the human user, respectively. Having a snack Living room Watching television Table 2: A comparison of conversational preferences by the type of Engaged in a desk activity interaction when the user was alone and when with few people around. Engaged in a conversation Tidying up Alone With people around Bed room Resting NI, GRT, SER 64% 79% Engaged in a desk activity TLK, CON 36% 21% Cleaning Kitchen Preparing a meal Having breakfast or few other known persons are present around. In the exper- iment, participants in a single setting were acquaintances. For instance, if there were few people around the user at the time to disturb the user. During the experiment, robot had a CON of the conversation, all these people were acquaintances but with its user and in the end, a survey was conducted to know were not related to each other. the actual preference of the user. Users were shown how the MIRob was remotely controlled by a human operator. type of conversations are categorized and were asked to select Robot responses were generated with respect to the user their preference at a similar occasion in the participant’s own response in each occasion. It is expected to assess the effect domestic environment despite the conversation he/she of considered factors in the surrounding upon human conver- already had with the robot. sational preferences, in both qualitative and quantitative man- The robot initiated a conversation despite the task of the ners. Therefore, the independent variables used in the study user, and user responses towards that interaction were were the task, area of the social environment (living room, recorded. Voice responses were monitored remotely by an bedroom, or kitchen), and whether few people were present operator without the knowledge of the user. Furthermore, a around the user or not. The type of interaction preferred by single participant was asked to perform all the tasks listed the user is used as the dependent variable in the analysis stage. in the experiment separately in different occasions. Each task As stated earlier, in the experiment, the same occasion was was performed twice: when the user was alone and when few analyzed under two categories: when only the user was present others are present. The second occasion replicates a typical and when there were few other people were around. An domestic or a social environment in which family members important fact to be considered in this case was that when Applied Bionics and Biomechanics 7 (a) (b) (c) Figure 6: An example scenario during the experiment is shown. (a) The user was in the kitchen, having a drink, (b) the robot approached user and initiated a conversation, and (c) interaction continued. few other people were present around, they were not involved in the interaction process except when the user was having a conversation with them. As the experiment was conducted by means of a WoZ, the robot operator monitored the robot towards a point closer to the user. Voice responses were gen- erated after the robot approached the user. Path planning and navigation of the robot were autonomous while tracking of the user and generation of voice responses were teleoper- ated by the human operator. Therefore, the operator instructed the robot where to approach and what to speak. Figure 7: A situation in which the user was having a drink and in None of the persons in the environment participated in the the surrounding, there was another human without an interaction conversation with the robot except the intended participant. with the user. Robot approached the user and initiated a As MIRob was monitored by a human operator, its voice conversation. responses were generated in accordance with the responses from the participant. The responses of the robot during the of interactions NI, GRT, and SER are categorized into a single experiment include only maintaining a socially interactive group for analysis because these types are preferred by distance between the robot and the user and voice. If any of humans in official situations and whenever there is little time the users does not respond the robot, the robot was for relaxation or friendly behavior. Therefore, TLK and instructed to leave without causing any distraction. In such CON, which fall under friendlier conversational preferences, a situation, the robot assumes that the user does not prefer are grouped together. An example scenario from the experi- to interact. ment is given in Figure 6 when the user was alone. A scenario Independent variables used in the experiment were the when there were people around is shown in Figure 7. In this task of the user, domestic area, and the presence of others situation, the user preferred a long conversation when she in the surrounding. The conversational preference was the was alone, and a service when there was a second person in dependent variable during analysis. The assumption made the kitchen. Such behavioral changes were recorded during during the study was that there is no significant difference the experiment. between the groups used for comparison purposes. Table 3 shows an ANOVA test performed on the same data for the comparison of percentage frequencies of each 4.3. Results of the Experiment. After the experiment, the con- type of conversational preference in each area of the social environment. The test was performed to analyze how the ten- versational preferences of users were analyzed using statisti- cal methods. The first question of interest was whether dency towards each type of conversational preference there is a difference in user responses depending on the num- changes when the domestic area changes. Here, living room, ber of people in the surrounding. Table 2 shows a compari- bed room, and the kitchen were used as living areas as men- son of the percentage frequency of each type of interaction tioned before. Percentage usage of the types of interaction is calculated and compared. First, test was implemented for for the two occasions: when the user was alone and when few people were around. This study was intended for all the the case when only the user was alone in the considered envi- tasks listed in Table 1. As seen from the results, there is an ronment, and the second test for the case when few others were present in the surrounding, in addition to the user. increase in demanding a service or limiting the conversation just for a greeting when few other people were present in the From the test, it was intended to find the differences in con- versational preferences when condition of the surrounding surrounding. As seen from this information, the demand for interaction types NI, GRT, and SER have been increased by with regard to the peer (whether the user was alone or there 15% when the number of people around the user has were few others around) was kept constant. Furthermore, it was expected to find whether there is a change in user behav- increased from zero to a few. In the same way, the tendency towards friendly conversations (TLK and CON) has been ior upon where the user is, despite whether he/she is alone or with few people around. reduced from 36% to 21%, i. e., by 15%. In this case, types 8 Applied Bionics and Biomechanics Table 3: ANOVA test for the comparison of percentage frequencies Table 4: t-test for the comparison among each type of interaction of each interaction type in each area of the social environment. when the user was alone and with few people around. Type of interaction T scores Alone With people (a) Mean 20.67 22.33 Alone Variance 210.33 16.33 Mean Variance Dof 2 NI Living room 20 107.39 0.852 Bedroom 20 127.99 t 4.302 Kitchen 20 286.66 Mean 8 12.67 Variance 7 16.33 ANOVA test SS DOF F p value GRT 0.034 Between groups 0 2 0 1 4.302 Within group 2088.22 12 Mean 35.33 44.33 Total 2088.22 14 Variance 82.33 54.33 SER 0.046 (b) t 4.302 With people around Mean 21.67 11.67 Mean Variance Variance 100.33 9.33 TLK Living room 20 116.73 0.131 Bedroom 20 309.39 t 4.302 Kitchen 20 263.53 Mean 14 8.67 Variance 9 8.33 ANOVA test CON SS DOF p value 0.246 Between groups 0.1333 2 0.00029 0.9997 t 4.302 Within group 2756.8 12 Total 2756.93 14 people were not directly involved with him/her, their pres- ence influenced the reactions of the user towards robot. This Table 4 shows the results of a t-test performed to test the could be explained using the theory of planned behavior [37]. deviation between the preference of each type of interaction A perceived behavioral control could be observed within the when alone and when surrounded by a few. Changes in user due to such changes in the surrounding. demand for each type of interaction in the said two occasions Most important and unexpected patterns in user behav- were analyzed without an involvement of other types of con- ior were demonstrated from the t-test shown in Table 4. versational preferences. Frequency of the type of interaction For all the conversational preferences except GRT and SER, in each domestic area was taken as data for the t-test. This p >0:05. Hence, the significance of the effect in the cases 1 explored unexpected tendencies in human behavior, and an and 2 becomes of interest. A probable reason for this is that, in-depth analysis of the results is given in the discussion. in almost all the occasions, NI was preferred, and the user Shown in Table 5 are two ANOVA tests performed on gave prominence to the task despite how many people were the same set of data to test the deviation between the conver- around. This was the same when the user preferred TLK sational preferences during the list of selected tasks while the and CON as well. In such situations, the user gave promi- user was alone and with one/few people around. The fre- nence for relaxation by means of conversation, rather than quency of using each conversational preference during these the task. An example was when the user was in a phone call tasks was calculated and analyzed for the deviations between or a desk activity. In such a situation, user will not prefer each group. Here, the groups were the conversational prefer- to be interacted. Hence, the conversational preference ences from NI to CON and the frequencies were listed becomes NI. If the user was having a snack, alone, in the according to the tasks listed in Table 1. In both the situations living room, he would prefer to have a long conversation in Table 5, when alone and when surrounded by few people, and will focus on the conversation without much consider- the F critical value was 2.539. ation about performing the task properly. As a whole, con- versational preferences at the two ends: NI and “having a 4.4. Observations and Discussion. From the results displayed friendly conversation” (TLK and CON) had no influence in Table 2, demand for conversational preferences NI, GRT, from the living area but middle interaction types (GRT and SER was decreased by 15% as the number of people and SER) had. For GRT and SER, where p =0:034 and around the user changed from “none” to “few.” It can be seen p =0:046, the null hypothesis could not be accepted. Hence, that the tendency of the user towards a friendly interaction it can be concluded that there exists a significant difference in reduced when there were people around. Even though these Applied Bionics and Biomechanics 9 Table 5: ANOVA test for the comparison of the frequencies of each conversational preference during each task. (a) Alone Groups Mean Variance NI 24.32 516.63 NI GRT SER TLK CON GRT 8.56 27.67 With people around 103 59 188 51 43 When alone 108 38 146 93 59 SER 32.88 317.20 TLK 20.95 256.49 Type of interaction preferred CON 13.29 102.87 When alone With people around ANOVA test SS DOF F p value Figure 8: A stacked graph drawn for the comparison of Between groups 4338.20 4 4.442 0.0035 conversational preferences with the two conditions: when the user is alone and when surrounded by few people. The type of Within groups 13429.51 55 interaction is plotted against the frequency of each type of Total 17767.71 59 interaction preferred in above two the occasions. (b) were preferred by the users mostly when they were alone. The percentage differences for these two types of interaction were With people around 28% and 16%. The highest percentage difference for these Groups Mean Variance two occasions was observed in TLK. A possible reason for NI 23.20 162.64 this is that TLK is the most flexible type of interaction which GRT 13.29 65.69 a user can have without getting disturbed to his/her task. In SER 42.34 193.02 the meantime, the user will get a chance to have a friendly interaction with the robot, without getting bored by the task TLK 11.49 25.40 or too involved in the task. CON 9.68 53.73 From the results shown in Table 3, behavioral changes observed when the user was alone, and when few people were ANOVA test SS DOF p value around were analyzed separately. From the first ANOVA test, a p value of 1 (≥0:05) and an F value of NI could be Between groups 8800.10 4 21.979 7.053E-11 observed for comparing conversational preferences within Within groups 5505.23 55 each social area: living room, bed room, and the kitchen. Total 14305.33 59 Therefore, the fact that “there is a significant difference in the conversational preferences with the social area when the user was performing a task alone” cannot be accepted. In the same way, from the second ANOVA test in Table 3, the conversational preference for GRT and SER, when the when few people were around, p value of 0.9997 ( ~ 1) and living area changes. Significant rises and drops in conversational preferences an F value of 0.00029 ( ~ 0). Hence, the fact that “there is a significant difference in conversational preferences when were observed with the change in the number of people around. This is demonstrated in Figure 8. When the overall the user was surrounded by a few people in the surrounding” also cannot be accepted. From the two tests, we could frequencies of NIs for all the tasks for both occasions are con- observe that there is no significant effect of the type of living sidered, there was a drop in “no interaction” preference when area upon conversational preference of a particular user but few people were present around the user. One possible reason his task. for this is that a user tend to take a service from the robot, on behalf of all the humans around. However, this drop was not According to the two ANOVA tests in Table 5, in both the cases, when the user was alone and was with one/few from a significant percentage. The inverse happened with people around, F values (4.442, 21.979) were larger than F “greeting”; the demand for GR was higher when few people were around the user. The reason for this is the human ten- critical (2.539). Hence, in both these cases, the null hypothe- sis can be rejected. Hence, the assumption that “there is no dency to hide the desire towards interaction and become inwardly in a social environment. Therefore, people became significant difference between each type of conversational preference during the selected set of tasks” was declined. more introvert with the presence of other humans. The expectancy of service increased when there were few people Therefore, it can be deduced that the preferences for NI to around. Therefore, a significant increase for SER was CON were significantly different when the given tasks were considered. Furthermore, in both the occasions, p values observed when the user situation changed from “alone” to “with few people around.” This increase was by 14%. As (0.035, 7.053E-11) were smaller than the alpha variable (0.05). This also suggests that the individual variables were TLK and CON are rather friendlier types of interaction, these Frequency (%) 10 Applied Bionics and Biomechanics NI GRT SER TLK CON NI GRT SER TLK CON NI GRT SER TLK CON NI GRT SER TLK CON NI GRT SER TLK CON NI GRT SER TLK CON Resting while sitting Reading while sitting Having a snack Watching television Desk activity Making a phone call Type of interaction NI TLK GRT CON SER Figure 9: This graph depicts how the users picked up conversational preferences during the selected tasks while the domestic area and people in the surrounding were kept constant. Here, the domestic area was the living room, and the user was alone in the area. statistically significant. During “with people” situation, the F interaction when most tasks were considered. This fact was value (21.979) was significantly larger than the F critical confirmed by the results shown in Figure 9. (2.539). Hence, the joint effect of all the variables together is larger than that when the user was “alone.” 5. Conclusions and Implications Another fact observed during the study was that the exis- tence of a significant difference in conversational preferences Continuation of a conversation while perceiving conversa- based on the task. This is examined in the chart in Figure 9. tional preferences of a user is an important aspect in In Figure 9, the frequency of the users who used each type human-robot interaction. In the paper, findings related to of interaction is plotted against each type of interaction while human conversational preferences from a WoZ experiment the user was engaged in the selected task. In all the occasions, are presented. Interaction was initiated in the form of a con- the domestic area was the living room, and the user was alone versation between the robot and the human. The length of a in the environment. Unlike previous experiments, here, we conversation was used as a mediator to monitor the user categorized conversational preferences based on the current preference for a short or long interaction with the robot. In task of the participant. We recorded the number of partici- this case, the conversational preference was used as a major pants who go for each conversational preference during the contributor to perceive human interest and attention towards activity. For example, 32% of the total participants preferred the robot while some factors in the environment or factors SER when they were “reading while being seated.” This is within the user change. According to the current researches, represented by the third (in green) column under “Reading the behavior of humans among acquaintances, their while sitting” in the graph shown in Figure 9. As seen from responses will be friendlier in the presence of family or rela- the chart, there were significant differences in user’s conver- tives (e.g., a domestic environment) and less friendly in the sational preferences when their task changed. For example, presence of strangers (e.g., a public space). Therefore, this few users have chosen NI while resting but many users have study can be used to find tendencies of humans in general chosen NI while making a phone call. In the two occasions, and to derive those in common encounters. the percentage frequency of users adopted NI was 11% and A teleoperated robot was used to perceive human situa- 76% which reflect a huge difference in adoption of NI during tion by means of conversational preferences when the above the two activities. As a whole, there was a considerable varia- factors were subjected to change. The experiment was tion in conversational preferences in the six tasks considered intended to reveal the relationship between internal (user- here. People were comfortable with only certain types of related) and external (environment-related) and Frequency (%) Applied Bionics and Biomechanics 11 number of parameters must be observed from the user and conversational preferences of humans. We considered three such factors: user’s task, people in the surrounding, and the his/her environment before the decision-making process of type of domestic area. During the study, we intended to see a robot. Therefore, this could not replicate all parts of the HHI (human-human interaction) into the HRI scenario. if these internal and external factors influence the conversa- tional preference of a person. We considered five conversa- These findings were based on the assumption that people tional preferences: no interaction, greeting, asking for a prefer the same rules of interaction with the robot as they do service, small talk, and long conversation, depending on the when interacting with humans. There can be certain cultures length of the conversation. Interesting facts regarding con- and social groups in which there are alterations in this fact [38]. Hence, such persons would react to robots in a different versational preferences based on the changes related to the user and the surrounding were revealed during the analysis manner. In addition, behavior adaptation is as important as of data. The findings of the study are expected to be used to behavior monitoring in such a scenario. Several other factors rebuild modern interaction mechanisms among humans which influence interaction such as the gender, previous and robots, so that the two conversants (human and robot) experience, and familiarity with the robot were not consid- ered within the context of this experiment. are motivated towards a sustaining conversation. Results show that there are considerable effects from factors in the surrounding and the user, upon the conversational prefer- 5.2. Implications for Design. Findings suggest that this evalu- ation offers better means of determining an appropriate con- ence of a user at that particular time. Moreover, despite age differences, these factors have become prominent in deciding versational preference based on several factors within the user and the environment. As users prefer their robots not conversational preferences during a particular moment. Furthermore, these findings can be made useful in develop- to interrupt their usual behavior, the first design guideline ing adaptive robotics systems which are expected to be used suggested from these findings is to respect the preferences of humans by simply following their concerns. These “con- in social environments. Although WoZ allowed us to prototype a domestic cerns” can be determined by the factors considered in the study. This “se user situation further acts as an etiquette human-robot scenario, the simulation process was con- nse” of strained by some realistic situations. As a result, we have for the robot to fit well in social environments. This can be implemented only a limited number of factors that affect presented as the second design guideline for social robots. human conversational preferences in a domestic human- The third design guideline is to extract information robot scenario. However, the system was capable to explore regarding the situation as much as possible. Considering a higher number of cues from the user and the environment novel tendencies in human behavior during human-robot interaction and successfully implemented the required con- increases the chance of an accurate perception of the situa- versation skills to make the process of interaction convenient tion. To perceive a number of such cues, the robot should and friendly. Furthermore, we believe that there will be other acquire visual and auditory sensory information for an ade- factors which could be influential in human conversational quate duration. This will be the forth design guideline for a situation-aware robot. preferences towards an interaction with a robot. For example, humans are more likely to accept a robot which will look and speak in the way a human does. Hence, patterns in speech, Data Availability appearance, and personality traits of the robot would also influence the acceptance of a robot. Therefore, evaluation of The data used to support the findings of this study may be such facts is important as well. In the future, it is expected released upon application to the corresponding author, Chapa to evaluate the personal characteristics of humans towards Sirithunge, who can be contacted at ra-chapa@uom.lk. conversational preferences. Moreover, present robots utilize limited capabilities in comparison to a human. Therefore, Conflicts of Interest the capabilities of the robot will eventually be improved in the future research. The authors declare that they have no conflicts of interest. Out of the three aspects: robot, user, and environment, only user and environment were evaluated in this study. In Acknowledgments the future, it is expected to evaluate the robot aspect as well, in order to make the human-robot interaction process much The authors would like to acknowledge the participants more lively and effective. who volunteered the study. This work was supported by the University of Moratuwa Senate Research Grant 5.1. Implications for Theory. Results suggest that there are SRC/LT/2018/20. factors in the environment and within the user itself, which affect user responses during a certain situation. Therefore, References the conceptual design of a robot’s intelligence must consider these factors before implementing its task-specific actions. [1] D. Sakamoto, Y. Sugiura, M. Inami, and T. Igarashi, “Graphi- Findings of the study were based on a limited number of cal instruction for home robots,” Computer, vol. 49, no. 7, tasks selected from the environment and the user itself. In a pp. 20–25, 2016. real-life scenario, this number will be much higher than the [2] B. Doughty, “Roller brush for surface cleaning robots,” 2016, number of factors considered here. Therefore, a maximum uS Patent 9, 326, 654. 12 Applied Bionics and Biomechanics interaction: a review and feasibility study,” Robotics, vol. 2, [3] J. R. Orejana, B. A. MacDonald, H. S. Ahn, K. Peri, and E. Broadbent, “Healthcare robots in homes of rural older no. 2, pp. 92–121, 2013. adults,” in International Conference on Social Robotics, [18] M. M. A. de Graaf, S. B. Allouch, and T. Klamer, “Sharing a life pp. 512–521, Springer, 2015. with Harvey: Exploring the acceptance of and relationship- [4] S. M. Rabbitt, A. E. Kazdin, and B. Scassellati, “Integrating building with a social robot,” Computers in Human Behavior, vol. 43, pp. 1–14, 2015. socially assistive robotics into mental healthcare interventions: applications and recommendations for expanded use,” Clinical [19] K. Dautenhahn, “Socially intelligent robots: dimensions of Psychology Review, vol. 35, pp. 35–46, 2015. human–robot interaction,” Philosophical Transactions of the Royal Society B: Biological Sciences, vol. 362, no. 1480, [5] M. Fridin and M. Belokopytov, “Acceptance of socially assis- tive humanoid robot by preschool and elementary school pp. 679–704, 2007. teachers,” Computers in Human Behavior, vol. 33, pp. 23–31, [20] T. Fong, I. Nourbakhsh, and K. Dautenhahn, “A survey of socially interactive robots,” Robotics and Autonomous Systems, [6] A. Yorita and N. Kubota, “Cognitive development in partner vol. 42, no. 3-4, pp. 143–166, 2003. robots for information support to elderly people,” IEEE Trans- [21] C. D. Kidd and C. Breazeal, “Robots at home: Understanding actions on Autonomous Mental Development, vol. 3, no. 1, long-term human-robot interaction,” in 2008 IEEE/RSJ Inter- pp. 64–73, 2011. national Conference on Intelligent Robots and Systems, pp. 3230–3235, Nice, France, September 2008. [7] R. Randell, J. Greenhalgh, J. Hindmarsh et al., “Integration of robotic surgery into routine practice and impacts on commu- [22] R. C. Arkin, M. Fujita, T. Takagi, and R. Hasegawa, “An etho- nication, collaboration, and decision making: a realist process logical and emotional basis for human-robot interaction,” evaluation protocol,” Implementation Science, vol. 9, no. 1, Robotics and Autonomous Systems, vol. 42, no. 3-4, pp. 191– p. 52, 2014. 201, 2003. [8] C. Ray, F. Mondada, and R. Siegwart, “What do people expect [23] B. Mutlu, T. Shiwa, T. Kanda, H. Ishiguro, and N. Hagita, from robots?,” in 2008 IEEE/RSJ International Conference on “Footing in human-robot conversations: how robots might Intelligent Robots and Systems, pp. 3816–3821, Nice, France, shape participant roles using gaze cues,” in Proceedings of the September 2008. 4th ACM/IEEE international conference on Human robot [9] S. Strohkorb and B. Scassellati, “Promoting collaboration with interaction - HRI '09, pp. 61–68, La Jolla, CA, USA, 2009. social robots,” in 2016 11th ACM/IEEE International Confer- [24] M. A. Salichs, I. P. Encinar, E. Salichs, Á. Castro-González, and ence on Human-Robot Interaction (HRI), pp. 639-640, Christ- M. Malfaz, “Study of scenarios and technical requirements of a church, New Zealand, March 2016. social assistive robot for alzheimers disease patients and their [10] G. Hoffman and C. Breazeal, “Robotic Partners Bodies and caregivers,” International Journal of Social Robotics, vol. 8, Minds: An Embodied Approach to Fluid Human-Robot Collab- no. 1, pp. 85–102, 2016. oration,” in AAAI Conference on Cognitive Robotics, 2006. [25] E. Law, V. Cai, Q. F. Liu et al., “A wizard-of-oz study of curiosity [11] M. Heerink, B. Vanderborght, J. Broekens, and J. Albó-Canals, in human-robot interaction,” in 2017 26th IEEE International “New friends: social robots in therapy and education,” Interna- Symposium on Robot and Human Interactive Communication tional Journal of Social Robotics, vol. 8, no. 4, pp. 443-444, (RO-MAN), pp. 607–614, Lisbon, Portugal, August 2017. [26] D. Hebesberger, T. Koertner, C. Gisinger, and J. Pripfl, “A [12] B. Robins, K. Dautenhahn, R. T. Boekhorst, and A. Billard, long-term autonomous robot at a care hospital: a mixed “Robotic assistants in therapy and education of children with methods study on social acceptance and experiences of staff autism: can a small humanoid robot help encourage social and older adults,” International Journal of Social Robotics, interaction skills?,” Universal Access in the Information Society, vol. 9, no. 3, pp. 417–429, 2017. vol. 4, no. 2, pp. 105–120, 2005. [27] J. Miehle, I. Bagci, W. Minker, and S. Ultes, A Social Compan- [13] M. N. Nicolescu and M. J. Mataric, “Learning and interacting ion and Conversational Partner for the Elderly, Springer Inter- national Publishing, Cham, 2019. in human-robot domains,” IEEE Transactions on Systems, man, and Cybernetics-part A: Systems and Humans, vol. 31, [28] T. Kanda, M. Shiomi, Z. Miyashita, H. Ishiguro, and N. Hagita, no. 5, pp. 419–430, 2001. “An affective guide robot in a shopping mall,” in Proceedings of [14] M. L. Traeger, S. Strohkorb Sebo, M. Jung, B. Scassellati, and the 4th ACM/IEEE international conference on Human robot N. A. Christakis, “Vulnerable robots positively shape human interaction - HRI '09, pp. 173–180, La Jolla, CA, USA, 2009. conversational dynamics in a human-robot team,” Proceedings [29] Y. Iwamura, M. Shiomi, T. Kanda, H. Ishiguro, and N. Hagita, of the National Academy of Sciences of the United States of “Do elderly people prefer a conversational humanoid as a America, vol. 117, no. 12, pp. 6370–6375, 2020. shopping assistant partner in supermarkets?,” in Proceedings [15] E. S. Short, E. C. Deng, D. Feil-Seifer, and M. J. Matarić, of the 6th international conference on Human-robot interaction “Understanding agency in interactions between ChildrenWith - HRI '11, pp. 449–457, EPFL, Lausanne, Switzerland, 2011. autism and socially assistive robots,” Journal of Human-Robot [30] H. P. C. Sirithunge, M. A. V. J. Muthugala, A. G. B. P. Jayase- Interaction, vol. 6, no. 3, pp. 21–47, 2017. kara, and D. P. Chandima, “Interpretation of interaction [16] T. Ozeki, T. Mouri, H. Sugiura, Y. Yano, and K. Miyosawa, demanding of a user based on nonverbal behavior in a domes- tic environment,” in 2017 IEEE International Conference on “Use of communication robots to converse with people suffer- ing from schizophrenia,” ROBOMECH Journal, vol. 7, no. 1, Fuzzy Systems (FUZZ-IEEE), pp. 1–8, Naples, Italy, July 2017. [31] L. D. Riek, “Wizard of oz studies in hri: a systematic review and [17] L. Tiberio, A. Cesta, and M. Olivetti Belardinelli, “Psychophys- new reporting guidelines,” Journal of Human-Robot Interac- iological methods to evaluate users response in human robot tion, vol. 1, no. 1, 2012. Applied Bionics and Biomechanics 13 [32] S. R. Klemmer, A. K. Sinha, J. Chen, J. A. Landay, N. Aboobaker, and A. Wang, “Suede: a wizard of oz prototyp- ing tool for speech user interfaces,” in Proceedings of the 13th annual ACM symposium on User interface software and tech- nology - UIST '00, pp. 1–10, San Diego, CA, USA, 2000. [33] C. L. Sidner, C. Lee, and N. Lesh, “The Role of Dialog in Human Robot Interaction,” in International Workshop on Language Understanding and Agents for Real World Interac- tion, Hokkaido University, Japan, 2003. [34] C. Sirithunge, A. B. P. Jayasekara, and D. P. Chandima, “Pro- active robots with the perception of nonverbal human behav- ior: A review,” IEEE Access, vol. 7, pp. 77308–77327, 2019. [35] I. Ajzen, “From intentions to actions: a theory of planned behavior,” in Action control, pp. 11–39, Springer, 1985. [36] B. H. Sheppard, J. Hartwick, and P. R. Warshaw, “The theory of reasoned action: a meta-analysis of past research with rec- ommendations for modifications and future research,” Journal of Consumer Research, vol. 15, no. 3, pp. 325–343, 1988. [37] I. Ajzen, “Perceived behavioral control, self-efficacy, locus of control, and the theory of planned Behavior1,” Journal of Applied Social Psychology, vol. 32, no. 4, pp. 665–683, 2002. [38] H. P. C. Sirithunge, M. A. V. J. Muthugala, A. G. B. P. Jayase- kara, and D. P. Chandima, “A wizard of oz study of human interest towards robot initiated human-robot interaction,” in 2018 27th IEEE International Symposium on Robot and Human Interactive Communication (RO-MAN), pp. 515–521, Nanjing, China, August 2018. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Applied Bionics and Biomechanics Hindawi Publishing Corporation

An Evaluation of Human Conversational Preferences in Social Human-Robot Interaction

Loading next page...
 
/lp/hindawi-publishing-corporation/an-evaluation-of-human-conversational-preferences-in-social-human-y3ngK1mr0w
Publisher
Hindawi Publishing Corporation
Copyright
Copyright © 2021 Chapa Sirithunge et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
ISSN
1176-2322
eISSN
1754-2103
DOI
10.1155/2021/3648479
Publisher site
See Article on Publisher Site

Abstract

Hindawi Applied Bionics and Biomechanics Volume 2021, Article ID 3648479, 13 pages https://doi.org/10.1155/2021/3648479 Research Article An Evaluation of Human Conversational Preferences in Social Human-Robot Interaction Chapa Sirithunge , A. G. Buddhika P. Jayasekara , and D. P. Chandima Intelligent Service Robotics Group, Department of Electrical Engineering, University of Moratuwa, Moratuwa 10400, Sri Lanka Correspondence should be addressed to Chapa Sirithunge; hpcschapa@gmail.com Received 8 January 2020; Revised 18 September 2020; Accepted 29 January 2021; Published 23 February 2021 Academic Editor: Francesca Cordella Copyright © 2021 Chapa Sirithunge et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. To generate context-aware behaviors in robots, robots are required to have a careful evaluation of its encounters with humans. Unwrapping emotional hints in observable cues in an encounter will improve a robot’s etiquettes in a social encounter. This article presents an extended human study conducted to examine how several factors in an encounter influence a person’s preferences upon an interaction at a particular moment. We analyzed the nature of conversation preferred by a user considering the type of conversation a robot could have with its user, having the interaction initiated by the robot itself. We took an effort to explore how such preferences differ as the factors present in the surrounding alter. A social robot equipped with the capability to initiate a conversation is deployed to conduct the study by means of a wizard-of-oz (WoZ) experiment. During this study, conversational preferences of users could vary from “no interaction at all” to a “long conversation.” We changed three factors in an encounter which can be different from each other in each circumstance: the audience or outsiders in the environment, user’s task, and the domestic area in which the interaction takes place. Conversational preferences of users within the abovementioned conditions were analyzed in a later stage, and critical observations are highlighted. Finally, implications that could be helpful in shaping future social human-robot encounters were derived from the analysis of the results. set of restrictions to abide by during an interaction with the 1. Introduction robot. Intelligence in initiating conversations at right occasions Acceptance of service robots in social environments has inspired many researchers to explore human tendencies is highly appreciated in achieving a robot’s context-aware behavior. Many users prefer to interact with robots, with when dealing with social robots. Conventional service robots speech [8], and the nature of speech must be friendlier and deployed in social environments are expected to support human-like. In other words, these robots are preferred to daily routine tasks such as cooking, cleaning, and taking care determine when to interact and when not to. During the of health [1–3], but modern assistive robots must also have cognitive skills to maintain a friendly and human-like inter- interaction, this natural behavior enhances the cohesion between robot and its nonexpert user. Such robotic systems action with the humans they daily meet [4]. For these robots, to be accepted by its user for a long duration, certain human- which can replicate complex human behavior in order to play the role of a close contact such as “friend” rather than a “ser- like qualities have to be embedded. Perceived sociability, cog- nitive skills, and adaptation are found to be the key factors vant” are being developed [9–13]. On the one hand, humans considered in long-term acceptance of a social robot [5]. Fur- prefer robots with at least some context awareness as well, in thermore, making right interaction decisions is equally addition to performing a predefined set of tasks. As a result, important in playing the role of a companion rather than robots can collaborate with people without disturbing them being just a service provider [6, 7]. These features ease when they are engaged in an activity. On the other hand, dealing with the robot without stressing out its user with a when robots have an instinct of how to co-op with the 2 Applied Bionics and Biomechanics situation, users do not need to stress themselves with a set of the conventions of conversation to connect with their human predefined behaviors that are perceivable by the robot. counterparts. In [18], the authors have investigated occasions Therefore, robotic systems, which adapt to the circumstances in which active collaboration between a robot and a human is in a certain encounter, are demanding in this era. Hence, required in service applications. Perceived connection between the human and the robot becomes effective, and social intelligence is an emerging requirement in human- the mutual interest grows when the two participants can robot interaction concerning social environments. Likeliness of the robot being accepted as a conversational understand the intentions of each other in a conversation; for instance, when to continue interaction and when to stop. partner depends on the environment as well as the current task of the user. The study in [14] ensures how the social As per this study, understanding the situation of a human is a demanding feature in robot’s acceptance within a sociable behavior of a robot improves the acceptance from its users human environment. as a companion in social human-robot domains. We investi- gate how the nature of interaction initiated by the robot Certain features and behaviors embodied in robots make an impact on people’s willingness to engage in at affects its acceptance by a human. Subsequently, a set of affect-based types of conversation were selected and imple- least a short interaction with the robot. The work explained in [19] has presented a set of social rules for robot behavior mented on a service robot in a simulated social environment (a “robotiquette”) that is comfortable and acceptable to in which few users were present or only a single user was present. This study is intended to find human tendencies humans. According to that, the conceptual space of HRI (human-robot interaction) studies expects a robot compan- towards interaction in different situations and hence will pro- vide means of engraving social skills into a robot’s behavior ion in a home environment to “do the right things” and “fulfil its tasks” in a manner that is acceptable and comfortable to before utilizing in human environments. To support this humans. Furthermore, real-time performance of the robot approach, we conducted a wizard-of-oz study to explore the nature of human conversation with the presence of a service which follows human social conventions and norms is more likely to be accepted for a long duration by humans [20]. robot when several factors in the environment vary. Factors which are more likely to have an influence upon the conver- In [21], a robot which is also an intelligent weight loss coach has been implemented. This makes an excellent exam- sational preference of humans are selected. The type of inter- ple for the situation perception embodied in the robot itself action preferred by the user was used as a mediator to perceive user situation and the level of interest towards the and hence has been exploited to reduce obesity. During this case, results show that the robot is accepted for a long-term interaction initiated by the robot. Responses observed during the study can be used to upgrade existing robots’ perception interaction by its users. However, this robot is not fully capa- of human behavior. Hence, this will allow a robot to be a ble to identify user behaviors which are not related to physi- cal health. But the fact that a higher social intelligence as well successful companion to the human without violating user expectations. as a greater acceptance from the user can be achieved by engraving abilities related to emotional and instinctive As humans establish emotional ties with whom they interact, this fact may remain the same for a human-robot behavior cannot be neglected in this scope [22]. Furthermore, interaction as well. Hence, the proactive, social means of there are many robotic systems to carry out a smooth conver- sation but the capability of these systems is limited to only engagement are expected from a robot in such a scenario [15]. One aspect in developing mechanisms to enhance social after the initiation of a conversation but not before the con- versation [23]. In [24], conversation was used as a part of intelligence in robots is to improve the user experience with such robots. A similar study was conducted to analyze con- assistance for Alzheimer patients in addition to therapy. This versational preferences of schizophrenia patients with robots study has identified robot’s knowledge of the location, patient’s history, type of disease, etc. and these parameters in [16]. Even so, only the context covers only patient-robot interaction, not human-robot interaction entirely. are important to decide the level of interaction between the patient and the robot. According to the survey by Lorenza et al. [17], the cogni- Law et al. [25] present a similar approach towards tive, emotional, and behavioral examination of human responses have an impact upon a robot’s behavior. Therefore, understanding one aspect in this regard. The authors have conducted a human study by means of a wizard-of-oz to measure this impact adequately beforehand, it is important for the robots to be cautious upon the factors which affect a experiment, to assess the level of curiosity aroused in humans when dealing with an assistive robot. Study con- human’s interaction decisions during a particular encounter. firmed the fact that the human curiosity considerably From this study, we intend to lay a justifiable basis to bring several such observable factors that can be used by a robot to changes when the intelligence of the robot is higher. More- over, it is found that the social acceptance of a service robot evaluate an encounter before robot-initiated interaction. We considered observable cues from humans as well as their sur- increases when a robot is able to perceive the very needs of a user and act accordingly [26]. Results of the human study roundings, which were likely to have an impact towards in [27] verify that humans prefer user adaptive dialogs in responses generated by a human in a certain instance. conversations even with a robot. The work explained in [28] is a promising example of the growing rapport 2. Related Work between humans and robots with such an intelligence. In order to participate in collaborations with people, robots Human further prefers the companionship build through interactive conversation between him and the robot; in must not only see and talk with people but also make use of Applied Bionics and Biomechanics 3 psychophysiological factors contribute majorly in deciding addition to the service purposes, most robots are intended for [29]. the level of interaction readiness within a human. Objects A method for attention estimation is proposed in [30]. and other humans in the surrounding, obstacles, etc. make Authors have used human pose and speeds of specific angu- the list of factors in the environment which have to be lar joints to identify the nonverbal interaction demanding of perceived by the robot. Figure 1 demonstrates this idea. a user. These two parameters were evaluated using a fuzzy logic-based mechanism to evaluate the interest level of a 3.2. Theory of Planned Behavior. Out of many psychologi- human towards a robot. However, these are not the only cal theories behind human tendencies, theory of planned parameters which define the nonverbal interaction demand- behavior lays a reasonable, yet justifiable basis for the dif- ing of a human. According to a review assessment performed ferences in human behavior under various circumstances in [31], wizard-of-oz is an effective mechanism to study in the environment. human tendencies based on nonverbal behavior. WoZ stud- According to the theory, one’s believes are linked to ies are effective in assessing such interactive sessions in a his/her behavior. This makes reasoned actions based on a short period without unnecessary preparations beforehand. restricted or controlled behavior. An individual’s intention An example scenario which evaluates speech-based interac- of a certain behavior at a specific time, a place, etc. is based tive interfaces is presented in [32]. In [33], Sidner et al. fur- on the regulations that humans follow by. These will take ther elaborate that the dialog features are important in three forms: behavioral, normative, and control. The theory accepting a robot for a long duration during human-robot of planned behavior comprises of six constructs that collec- interaction. tively present actual control of a person over the behavior The above discussed systems access a limited number of [35]. These constructs are stated briefly as follows. cues from its users and their surroundings before making interactive decisions. Still, there are several other factors to (1) Attitudes—degree to which the individual has a be considered before generating responses in a human- favorable or an unfavorable evaluation upon the robot social encounter [34]. We selected several such factors behavior of interest which are likely to have an impact upon user responses dur- ing an encounter and investigated whether these factors (2) Behavioral intention—motivational factors that actually make such an impact. Hence, the findings of the influence the behavior study can be used to improve the conceptual basis of reason- (3) Subjective norms—belief about whether behavior will ing in future social robots. be approved by peers and people of importance In our work, the importance of understanding user situ- ation is evaluated using such a WoZ approach. Factors used (4) Social norms—customary behavior in a group of to define user situation were the activity or the current task people belonging to a cultural context of the user, number of people around the user, and the type (5) Perceived power—the behavioral control over these of area of the house. Conversational preferences when these factors that may facilitate or impede performance factors change were analyzed with the help of a domestic ser- vice robot platform placed in a simulated social environment. (6) Perceived behavioral control—person’s perception of Since our work is based on verbal human behavior, a ease or the difficulty in performing the action WoZ experiment will have the capability to explore unex- pected tendencies in human behavior prior to an interaction. This concept can be related to human-robot scenario as In this work, we tried to investigate human behaviors that follows. Due to the factors in the environment, user’s percep- can be used as cues for a robot to perceive user situation prior tion of the environment or the surrounding may subject to to an interaction. It is expected that these findings will help change, depending on his/her beliefs. Hence, the reaction improve social intelligence of a social robot for the purpose towards robot’s conversations may change in different of caretaking and simultaneously providing emotional scenarios. Three factors which are most likely to affect the support through interaction. user response are considered in this study. These factors are selected from user and environment aspects. The purpose of this is to evaluate the human behavior during these situa- 3. Theoretical Approach tions. These factors are listed below. 3.1. Robot’s Perception of the Environment. A situation between a human and a robot consists of the robot itself, (i) Task of the user—e.g., having a snack, cleaning, and the user (human), and the environment (objects and space) engaged in a desk activity around the robot and the user. When the robot intends to (ii) People in the surrounding—alone or surrounded by perceive such a situation, it first has to identify interactive few people factors within itself, the environment, and the user. Factors within the robot itself include the dialog patterns the robot (iii) Type of area in the domestic/social environment—liv- generates, maintaining an interactive distance in between, ing room, bed room, or kitchen and displaying appropriate behavior, etc. Factors within the user will be numerous, but emotions, social norms, beliefs, These three factors contribute to evaluate mainly atti- personality traits, user’s activity at that moment, and other tudes, subjective norms, and perceived behavioral control 4 Applied Bionics and Biomechanics Perception criteria Example scenarios encountered in a social environment with the presence of humans and robots as perceived by the above the- 1.Self (robot) ories are given in Figures 2 and 3. In occasion 1 (Figure 2), the 2.User (human) Bed room user gave priority to an interaction with the robot, but in occasion 3.Environment 2 (Figure 3), the user gave priority to her current activity. In both User the occasions, various factors within the environment and the user itself will affect her response. Robot 4. Experiment Living room 4.1. Setting and the Research Platform. The experiment was conducted in a simulated social environment in the labora- tory. Participants were students, nonacademic staff members of the university, and some outsiders in the age range 19-58 (Mean-28.45, SD-9.02) who volunteered the study. There were 37 participants, and they were in good health condition Figure 1: An example domestic environment is shown. In this without any physical defects which will alter their reactions scenario, the user is involved in a desk activity in the bed room. In during the study. More than half of the participants did not order to understand the whole situation, the robot has to be have a technical background in education, majors, or knowledgeable on three aspects: itself, the user, and the research related to Engineering. The gender of the user was surrounding environment. Factors related to these three aspects are marked as 1,2, and 3, respectively. not included within the scope of this study. Upon arrival, the users were given instructions regarding the tasks they should complete but they were not aware of the fact that they out of the six constructs of the theory of controlled behavior. are intended to talk to the robot but they are instructed to It is assumed that the type of interaction preferred by the user respond towards the robot if the robot initiates an interac- change when these factors change. We evaluated this fact tion. They were not knowledgeable about the exact intention through the human study conducted in the form of the of the experiment because that will cause a bias response WoZ experiment. from users towards the robot. Hence, the participants were An application of the two theories: occasion 1 (a), user instructed to perform a given activity in the way they are used was working and has no idea about the presence of the robot, to perform that before. (b) notices the presence of the robot as it moves and as a The experiment was conducted using a service robot result, the user looks at the robot, and (c) stops the work called MIRob. The robot is visually and verbally capable and gives attention to the robot while it approaches the user. and has the ability to approach a user, make a conversation, An application of the two theories: occasion 2 (a), user and handle objects. This is a Pioneer 3DX MobileRobots was working and has no idea about the presence of the robot, platform equipped with a Cyton Gamma 300 manipulator (b) notices the presence of the robot as it moves and looks at and a Kinect camera for vision. The maps required to navi- the robot, (c) user averts her gaze and give attention to the gate around were created with Mapper3 Basic software. The work, and (d) engage in the work again. platform is equipped with a microphone and a speaker to listen to and respond its users. This platform is shown in 3.3. Theory of Reasoned Action. The theory explains that Figure 4. there is a relationship between one’s attitudes and actions [36]. Hence, the theory is used to predict the behavior of a 4.2. Procedure. The selected user was allowed to engage in a human in a particular scenario, based on the preexisted atti- certain task, and the robot was allowed to approach the user tudes and behavioral intentions of that individual. That indi- to initiate a conversation with him/her. The user was advised vidual’s expectations upon the outcomes of a behavior to complete a certain task and if the robot talks to him/her, to controls his/her decision to adopt that behavior. This fact is talk back. The set of tasks to be performed by the users was deployed in exploring the tendencies in human behavior in predefined. There were separate lists of tasks to be performed the presence of the robot used in the WoZ study. In such a in the living room, bed room, and kitchen. The participant or situation, there are few stages which a user goes through; the user was knowledgeable on the tasks that are to be per- for instance, noticing the presence of the robot, responding formed in each living area. The tasks were selected so that towards robot which approaches towards him/her, initiating there will be at least three tasks performed in each area. These a conversation with the robot, or responding to a conversa- tasks are the most common to that particular social or tion initiated by the robot can be stated as the usual stages domestic environment, and few tasks selected for the study of interaction in such a situation. The user’s conversational are listed in Table 1. While the user was engaged in a task, preferences might change according to his/her attitudes, the robot was remotely guided towards him/her and was beliefs, and expectations in such an instance. These attitudes, allowed to initiate an interaction in the ways given below. beliefs, and expectations may subject to change depending on This set of experiments was conducted over a period of 7 the factors present in the environment, and this fact is going days, so that the participants had enough time in between to be evaluated through this study. tasks. This prevented participants getting exhausted and Applied Bionics and Biomechanics 5 (a) (b) (c) Figure 2: An application of the two theories: occasion 1 (a), user was working and has no idea about the presence of the robot, (b) notices the presence of the robot as it moves and as a result, the user looks at the robot, and (c) stops the work and gives attention to the robot while it approaches the user. (a) (b) (c) (d) Figure 3: An application of the two theories: occasion 2 (a), user was working and has no idea about the presence of the robot, (b) notices the presence of the robot as it moves and looks at the robot, (c) user averts her gaze and give attention to the work, and (d) engage in the work again. hence generating biased, involuntary responses towards the CON-Long conversation. robot. People in the surrounding were not participants but How a conversation is categorized into these types is one or two members from the set of experimenters. And all shown in Figure 5. As the conversation extends, the type of 37 users participated in the experiments at least for three dif- conversation shifts from NI to a CON. The robot will not talk ferent tasks. As the number of participants was 37, each con- to its user during NI. In GR, the robot will only greet the per- ducted 12 tasks, 2 times (alone and with the presence of a few son and navigate away. The greeting will just be a single sen- tence saying “good morning,”“hey,”“hello,” etc. In SER, the others in the surrounding), the experiment was conducted 888 times throughout a week. We kept gaps in between robot will ask to deliver something for the user, as an assis- experiments to avoid users repeating the same response over tance to his/her current task. This will be approximately four and over by practice. The map of the environment was prede- sentences maximum in the entire conversation. In the TLK, fined in the simulation. Therefore, the robot navigated to the the robot will say a few additional sentences other than greet- target positions and its orientation which were defined by the ing and sometimes will ask if the user wants something. Such operator. In this scenario, the target position of the robot was a conversation consisted of about 5-7 sentences. All the con- a point within the interactive area near the user. For the ease versations longer than that were considered as CON. Such of future referencing, these types of conversational prefer- conversations cover a broader scope of topics as well as these ences are abbreviated as follows. existed for a longer duration. Therefore, the duration of the NI-No interaction conversation depended on the type of conversation robot GRT-Greeting had with a user. In all the occasions, the robot stopped con- SER-Asking to deliver a service tinuing the conversation depending on the curiosity of the TLK-Small talk user to engage with the robot or when the conversation seems 6 Applied Bionics and Biomechanics Conversation begins R: Good morning! H: A very good morning! Greeting R: You want me to get something for you? H: No, Thank you. Asking to deliver a service R: Okay. Are you tired? H: No, I’m just relaxing R: Okay. Small talk R: Is there something you want me to talk about? H: Hmm, There’s nothing specific, but I’d like a conversation. R: May I play a song? H: I would rather hear a poem. R: Good choice. A poem about nature? H: That’ll be great. Go on. Figure 4: Service Robot platform used in the experiment: MIRob. R: Have a nice day! H: You too! Long conversation Table 1: Some of the tasks selected for the study. Conversation ends Living area Task Figure 5: The nature and the length of conversations determine the Resting while sitting “type of conversation” existed at a certain occasion. Here “R” and Reading while sitting “H” represent the robot and the human user, respectively. Having a snack Living room Watching television Table 2: A comparison of conversational preferences by the type of Engaged in a desk activity interaction when the user was alone and when with few people around. Engaged in a conversation Tidying up Alone With people around Bed room Resting NI, GRT, SER 64% 79% Engaged in a desk activity TLK, CON 36% 21% Cleaning Kitchen Preparing a meal Having breakfast or few other known persons are present around. In the exper- iment, participants in a single setting were acquaintances. For instance, if there were few people around the user at the time to disturb the user. During the experiment, robot had a CON of the conversation, all these people were acquaintances but with its user and in the end, a survey was conducted to know were not related to each other. the actual preference of the user. Users were shown how the MIRob was remotely controlled by a human operator. type of conversations are categorized and were asked to select Robot responses were generated with respect to the user their preference at a similar occasion in the participant’s own response in each occasion. It is expected to assess the effect domestic environment despite the conversation he/she of considered factors in the surrounding upon human conver- already had with the robot. sational preferences, in both qualitative and quantitative man- The robot initiated a conversation despite the task of the ners. Therefore, the independent variables used in the study user, and user responses towards that interaction were were the task, area of the social environment (living room, recorded. Voice responses were monitored remotely by an bedroom, or kitchen), and whether few people were present operator without the knowledge of the user. Furthermore, a around the user or not. The type of interaction preferred by single participant was asked to perform all the tasks listed the user is used as the dependent variable in the analysis stage. in the experiment separately in different occasions. Each task As stated earlier, in the experiment, the same occasion was was performed twice: when the user was alone and when few analyzed under two categories: when only the user was present others are present. The second occasion replicates a typical and when there were few other people were around. An domestic or a social environment in which family members important fact to be considered in this case was that when Applied Bionics and Biomechanics 7 (a) (b) (c) Figure 6: An example scenario during the experiment is shown. (a) The user was in the kitchen, having a drink, (b) the robot approached user and initiated a conversation, and (c) interaction continued. few other people were present around, they were not involved in the interaction process except when the user was having a conversation with them. As the experiment was conducted by means of a WoZ, the robot operator monitored the robot towards a point closer to the user. Voice responses were gen- erated after the robot approached the user. Path planning and navigation of the robot were autonomous while tracking of the user and generation of voice responses were teleoper- ated by the human operator. Therefore, the operator instructed the robot where to approach and what to speak. Figure 7: A situation in which the user was having a drink and in None of the persons in the environment participated in the the surrounding, there was another human without an interaction conversation with the robot except the intended participant. with the user. Robot approached the user and initiated a As MIRob was monitored by a human operator, its voice conversation. responses were generated in accordance with the responses from the participant. The responses of the robot during the of interactions NI, GRT, and SER are categorized into a single experiment include only maintaining a socially interactive group for analysis because these types are preferred by distance between the robot and the user and voice. If any of humans in official situations and whenever there is little time the users does not respond the robot, the robot was for relaxation or friendly behavior. Therefore, TLK and instructed to leave without causing any distraction. In such CON, which fall under friendlier conversational preferences, a situation, the robot assumes that the user does not prefer are grouped together. An example scenario from the experi- to interact. ment is given in Figure 6 when the user was alone. A scenario Independent variables used in the experiment were the when there were people around is shown in Figure 7. In this task of the user, domestic area, and the presence of others situation, the user preferred a long conversation when she in the surrounding. The conversational preference was the was alone, and a service when there was a second person in dependent variable during analysis. The assumption made the kitchen. Such behavioral changes were recorded during during the study was that there is no significant difference the experiment. between the groups used for comparison purposes. Table 3 shows an ANOVA test performed on the same data for the comparison of percentage frequencies of each 4.3. Results of the Experiment. After the experiment, the con- type of conversational preference in each area of the social environment. The test was performed to analyze how the ten- versational preferences of users were analyzed using statisti- cal methods. The first question of interest was whether dency towards each type of conversational preference there is a difference in user responses depending on the num- changes when the domestic area changes. Here, living room, ber of people in the surrounding. Table 2 shows a compari- bed room, and the kitchen were used as living areas as men- son of the percentage frequency of each type of interaction tioned before. Percentage usage of the types of interaction is calculated and compared. First, test was implemented for for the two occasions: when the user was alone and when few people were around. This study was intended for all the the case when only the user was alone in the considered envi- tasks listed in Table 1. As seen from the results, there is an ronment, and the second test for the case when few others were present in the surrounding, in addition to the user. increase in demanding a service or limiting the conversation just for a greeting when few other people were present in the From the test, it was intended to find the differences in con- versational preferences when condition of the surrounding surrounding. As seen from this information, the demand for interaction types NI, GRT, and SER have been increased by with regard to the peer (whether the user was alone or there 15% when the number of people around the user has were few others around) was kept constant. Furthermore, it was expected to find whether there is a change in user behav- increased from zero to a few. In the same way, the tendency towards friendly conversations (TLK and CON) has been ior upon where the user is, despite whether he/she is alone or with few people around. reduced from 36% to 21%, i. e., by 15%. In this case, types 8 Applied Bionics and Biomechanics Table 3: ANOVA test for the comparison of percentage frequencies Table 4: t-test for the comparison among each type of interaction of each interaction type in each area of the social environment. when the user was alone and with few people around. Type of interaction T scores Alone With people (a) Mean 20.67 22.33 Alone Variance 210.33 16.33 Mean Variance Dof 2 NI Living room 20 107.39 0.852 Bedroom 20 127.99 t 4.302 Kitchen 20 286.66 Mean 8 12.67 Variance 7 16.33 ANOVA test SS DOF F p value GRT 0.034 Between groups 0 2 0 1 4.302 Within group 2088.22 12 Mean 35.33 44.33 Total 2088.22 14 Variance 82.33 54.33 SER 0.046 (b) t 4.302 With people around Mean 21.67 11.67 Mean Variance Variance 100.33 9.33 TLK Living room 20 116.73 0.131 Bedroom 20 309.39 t 4.302 Kitchen 20 263.53 Mean 14 8.67 Variance 9 8.33 ANOVA test CON SS DOF p value 0.246 Between groups 0.1333 2 0.00029 0.9997 t 4.302 Within group 2756.8 12 Total 2756.93 14 people were not directly involved with him/her, their pres- ence influenced the reactions of the user towards robot. This Table 4 shows the results of a t-test performed to test the could be explained using the theory of planned behavior [37]. deviation between the preference of each type of interaction A perceived behavioral control could be observed within the when alone and when surrounded by a few. Changes in user due to such changes in the surrounding. demand for each type of interaction in the said two occasions Most important and unexpected patterns in user behav- were analyzed without an involvement of other types of con- ior were demonstrated from the t-test shown in Table 4. versational preferences. Frequency of the type of interaction For all the conversational preferences except GRT and SER, in each domestic area was taken as data for the t-test. This p >0:05. Hence, the significance of the effect in the cases 1 explored unexpected tendencies in human behavior, and an and 2 becomes of interest. A probable reason for this is that, in-depth analysis of the results is given in the discussion. in almost all the occasions, NI was preferred, and the user Shown in Table 5 are two ANOVA tests performed on gave prominence to the task despite how many people were the same set of data to test the deviation between the conver- around. This was the same when the user preferred TLK sational preferences during the list of selected tasks while the and CON as well. In such situations, the user gave promi- user was alone and with one/few people around. The fre- nence for relaxation by means of conversation, rather than quency of using each conversational preference during these the task. An example was when the user was in a phone call tasks was calculated and analyzed for the deviations between or a desk activity. In such a situation, user will not prefer each group. Here, the groups were the conversational prefer- to be interacted. Hence, the conversational preference ences from NI to CON and the frequencies were listed becomes NI. If the user was having a snack, alone, in the according to the tasks listed in Table 1. In both the situations living room, he would prefer to have a long conversation in Table 5, when alone and when surrounded by few people, and will focus on the conversation without much consider- the F critical value was 2.539. ation about performing the task properly. As a whole, con- versational preferences at the two ends: NI and “having a 4.4. Observations and Discussion. From the results displayed friendly conversation” (TLK and CON) had no influence in Table 2, demand for conversational preferences NI, GRT, from the living area but middle interaction types (GRT and SER was decreased by 15% as the number of people and SER) had. For GRT and SER, where p =0:034 and around the user changed from “none” to “few.” It can be seen p =0:046, the null hypothesis could not be accepted. Hence, that the tendency of the user towards a friendly interaction it can be concluded that there exists a significant difference in reduced when there were people around. Even though these Applied Bionics and Biomechanics 9 Table 5: ANOVA test for the comparison of the frequencies of each conversational preference during each task. (a) Alone Groups Mean Variance NI 24.32 516.63 NI GRT SER TLK CON GRT 8.56 27.67 With people around 103 59 188 51 43 When alone 108 38 146 93 59 SER 32.88 317.20 TLK 20.95 256.49 Type of interaction preferred CON 13.29 102.87 When alone With people around ANOVA test SS DOF F p value Figure 8: A stacked graph drawn for the comparison of Between groups 4338.20 4 4.442 0.0035 conversational preferences with the two conditions: when the user is alone and when surrounded by few people. The type of Within groups 13429.51 55 interaction is plotted against the frequency of each type of Total 17767.71 59 interaction preferred in above two the occasions. (b) were preferred by the users mostly when they were alone. The percentage differences for these two types of interaction were With people around 28% and 16%. The highest percentage difference for these Groups Mean Variance two occasions was observed in TLK. A possible reason for NI 23.20 162.64 this is that TLK is the most flexible type of interaction which GRT 13.29 65.69 a user can have without getting disturbed to his/her task. In SER 42.34 193.02 the meantime, the user will get a chance to have a friendly interaction with the robot, without getting bored by the task TLK 11.49 25.40 or too involved in the task. CON 9.68 53.73 From the results shown in Table 3, behavioral changes observed when the user was alone, and when few people were ANOVA test SS DOF p value around were analyzed separately. From the first ANOVA test, a p value of 1 (≥0:05) and an F value of NI could be Between groups 8800.10 4 21.979 7.053E-11 observed for comparing conversational preferences within Within groups 5505.23 55 each social area: living room, bed room, and the kitchen. Total 14305.33 59 Therefore, the fact that “there is a significant difference in the conversational preferences with the social area when the user was performing a task alone” cannot be accepted. In the same way, from the second ANOVA test in Table 3, the conversational preference for GRT and SER, when the when few people were around, p value of 0.9997 ( ~ 1) and living area changes. Significant rises and drops in conversational preferences an F value of 0.00029 ( ~ 0). Hence, the fact that “there is a significant difference in conversational preferences when were observed with the change in the number of people around. This is demonstrated in Figure 8. When the overall the user was surrounded by a few people in the surrounding” also cannot be accepted. From the two tests, we could frequencies of NIs for all the tasks for both occasions are con- observe that there is no significant effect of the type of living sidered, there was a drop in “no interaction” preference when area upon conversational preference of a particular user but few people were present around the user. One possible reason his task. for this is that a user tend to take a service from the robot, on behalf of all the humans around. However, this drop was not According to the two ANOVA tests in Table 5, in both the cases, when the user was alone and was with one/few from a significant percentage. The inverse happened with people around, F values (4.442, 21.979) were larger than F “greeting”; the demand for GR was higher when few people were around the user. The reason for this is the human ten- critical (2.539). Hence, in both these cases, the null hypothe- sis can be rejected. Hence, the assumption that “there is no dency to hide the desire towards interaction and become inwardly in a social environment. Therefore, people became significant difference between each type of conversational preference during the selected set of tasks” was declined. more introvert with the presence of other humans. The expectancy of service increased when there were few people Therefore, it can be deduced that the preferences for NI to around. Therefore, a significant increase for SER was CON were significantly different when the given tasks were considered. Furthermore, in both the occasions, p values observed when the user situation changed from “alone” to “with few people around.” This increase was by 14%. As (0.035, 7.053E-11) were smaller than the alpha variable (0.05). This also suggests that the individual variables were TLK and CON are rather friendlier types of interaction, these Frequency (%) 10 Applied Bionics and Biomechanics NI GRT SER TLK CON NI GRT SER TLK CON NI GRT SER TLK CON NI GRT SER TLK CON NI GRT SER TLK CON NI GRT SER TLK CON Resting while sitting Reading while sitting Having a snack Watching television Desk activity Making a phone call Type of interaction NI TLK GRT CON SER Figure 9: This graph depicts how the users picked up conversational preferences during the selected tasks while the domestic area and people in the surrounding were kept constant. Here, the domestic area was the living room, and the user was alone in the area. statistically significant. During “with people” situation, the F interaction when most tasks were considered. This fact was value (21.979) was significantly larger than the F critical confirmed by the results shown in Figure 9. (2.539). Hence, the joint effect of all the variables together is larger than that when the user was “alone.” 5. Conclusions and Implications Another fact observed during the study was that the exis- tence of a significant difference in conversational preferences Continuation of a conversation while perceiving conversa- based on the task. This is examined in the chart in Figure 9. tional preferences of a user is an important aspect in In Figure 9, the frequency of the users who used each type human-robot interaction. In the paper, findings related to of interaction is plotted against each type of interaction while human conversational preferences from a WoZ experiment the user was engaged in the selected task. In all the occasions, are presented. Interaction was initiated in the form of a con- the domestic area was the living room, and the user was alone versation between the robot and the human. The length of a in the environment. Unlike previous experiments, here, we conversation was used as a mediator to monitor the user categorized conversational preferences based on the current preference for a short or long interaction with the robot. In task of the participant. We recorded the number of partici- this case, the conversational preference was used as a major pants who go for each conversational preference during the contributor to perceive human interest and attention towards activity. For example, 32% of the total participants preferred the robot while some factors in the environment or factors SER when they were “reading while being seated.” This is within the user change. According to the current researches, represented by the third (in green) column under “Reading the behavior of humans among acquaintances, their while sitting” in the graph shown in Figure 9. As seen from responses will be friendlier in the presence of family or rela- the chart, there were significant differences in user’s conver- tives (e.g., a domestic environment) and less friendly in the sational preferences when their task changed. For example, presence of strangers (e.g., a public space). Therefore, this few users have chosen NI while resting but many users have study can be used to find tendencies of humans in general chosen NI while making a phone call. In the two occasions, and to derive those in common encounters. the percentage frequency of users adopted NI was 11% and A teleoperated robot was used to perceive human situa- 76% which reflect a huge difference in adoption of NI during tion by means of conversational preferences when the above the two activities. As a whole, there was a considerable varia- factors were subjected to change. The experiment was tion in conversational preferences in the six tasks considered intended to reveal the relationship between internal (user- here. People were comfortable with only certain types of related) and external (environment-related) and Frequency (%) Applied Bionics and Biomechanics 11 number of parameters must be observed from the user and conversational preferences of humans. We considered three such factors: user’s task, people in the surrounding, and the his/her environment before the decision-making process of type of domestic area. During the study, we intended to see a robot. Therefore, this could not replicate all parts of the HHI (human-human interaction) into the HRI scenario. if these internal and external factors influence the conversa- tional preference of a person. We considered five conversa- These findings were based on the assumption that people tional preferences: no interaction, greeting, asking for a prefer the same rules of interaction with the robot as they do service, small talk, and long conversation, depending on the when interacting with humans. There can be certain cultures length of the conversation. Interesting facts regarding con- and social groups in which there are alterations in this fact [38]. Hence, such persons would react to robots in a different versational preferences based on the changes related to the user and the surrounding were revealed during the analysis manner. In addition, behavior adaptation is as important as of data. The findings of the study are expected to be used to behavior monitoring in such a scenario. Several other factors rebuild modern interaction mechanisms among humans which influence interaction such as the gender, previous and robots, so that the two conversants (human and robot) experience, and familiarity with the robot were not consid- ered within the context of this experiment. are motivated towards a sustaining conversation. Results show that there are considerable effects from factors in the surrounding and the user, upon the conversational prefer- 5.2. Implications for Design. Findings suggest that this evalu- ation offers better means of determining an appropriate con- ence of a user at that particular time. Moreover, despite age differences, these factors have become prominent in deciding versational preference based on several factors within the user and the environment. As users prefer their robots not conversational preferences during a particular moment. Furthermore, these findings can be made useful in develop- to interrupt their usual behavior, the first design guideline ing adaptive robotics systems which are expected to be used suggested from these findings is to respect the preferences of humans by simply following their concerns. These “con- in social environments. Although WoZ allowed us to prototype a domestic cerns” can be determined by the factors considered in the study. This “se user situation further acts as an etiquette human-robot scenario, the simulation process was con- nse” of strained by some realistic situations. As a result, we have for the robot to fit well in social environments. This can be implemented only a limited number of factors that affect presented as the second design guideline for social robots. human conversational preferences in a domestic human- The third design guideline is to extract information robot scenario. However, the system was capable to explore regarding the situation as much as possible. Considering a higher number of cues from the user and the environment novel tendencies in human behavior during human-robot interaction and successfully implemented the required con- increases the chance of an accurate perception of the situa- versation skills to make the process of interaction convenient tion. To perceive a number of such cues, the robot should and friendly. Furthermore, we believe that there will be other acquire visual and auditory sensory information for an ade- factors which could be influential in human conversational quate duration. This will be the forth design guideline for a situation-aware robot. preferences towards an interaction with a robot. For example, humans are more likely to accept a robot which will look and speak in the way a human does. Hence, patterns in speech, Data Availability appearance, and personality traits of the robot would also influence the acceptance of a robot. Therefore, evaluation of The data used to support the findings of this study may be such facts is important as well. In the future, it is expected released upon application to the corresponding author, Chapa to evaluate the personal characteristics of humans towards Sirithunge, who can be contacted at ra-chapa@uom.lk. conversational preferences. Moreover, present robots utilize limited capabilities in comparison to a human. Therefore, Conflicts of Interest the capabilities of the robot will eventually be improved in the future research. The authors declare that they have no conflicts of interest. Out of the three aspects: robot, user, and environment, only user and environment were evaluated in this study. In Acknowledgments the future, it is expected to evaluate the robot aspect as well, in order to make the human-robot interaction process much The authors would like to acknowledge the participants more lively and effective. who volunteered the study. This work was supported by the University of Moratuwa Senate Research Grant 5.1. Implications for Theory. Results suggest that there are SRC/LT/2018/20. factors in the environment and within the user itself, which affect user responses during a certain situation. Therefore, References the conceptual design of a robot’s intelligence must consider these factors before implementing its task-specific actions. [1] D. Sakamoto, Y. Sugiura, M. Inami, and T. Igarashi, “Graphi- Findings of the study were based on a limited number of cal instruction for home robots,” Computer, vol. 49, no. 7, tasks selected from the environment and the user itself. In a pp. 20–25, 2016. real-life scenario, this number will be much higher than the [2] B. Doughty, “Roller brush for surface cleaning robots,” 2016, number of factors considered here. Therefore, a maximum uS Patent 9, 326, 654. 12 Applied Bionics and Biomechanics interaction: a review and feasibility study,” Robotics, vol. 2, [3] J. R. Orejana, B. A. MacDonald, H. S. Ahn, K. Peri, and E. Broadbent, “Healthcare robots in homes of rural older no. 2, pp. 92–121, 2013. adults,” in International Conference on Social Robotics, [18] M. M. A. de Graaf, S. B. Allouch, and T. Klamer, “Sharing a life pp. 512–521, Springer, 2015. with Harvey: Exploring the acceptance of and relationship- [4] S. M. Rabbitt, A. E. Kazdin, and B. Scassellati, “Integrating building with a social robot,” Computers in Human Behavior, vol. 43, pp. 1–14, 2015. socially assistive robotics into mental healthcare interventions: applications and recommendations for expanded use,” Clinical [19] K. Dautenhahn, “Socially intelligent robots: dimensions of Psychology Review, vol. 35, pp. 35–46, 2015. human–robot interaction,” Philosophical Transactions of the Royal Society B: Biological Sciences, vol. 362, no. 1480, [5] M. Fridin and M. Belokopytov, “Acceptance of socially assis- tive humanoid robot by preschool and elementary school pp. 679–704, 2007. teachers,” Computers in Human Behavior, vol. 33, pp. 23–31, [20] T. Fong, I. Nourbakhsh, and K. Dautenhahn, “A survey of socially interactive robots,” Robotics and Autonomous Systems, [6] A. Yorita and N. Kubota, “Cognitive development in partner vol. 42, no. 3-4, pp. 143–166, 2003. robots for information support to elderly people,” IEEE Trans- [21] C. D. Kidd and C. Breazeal, “Robots at home: Understanding actions on Autonomous Mental Development, vol. 3, no. 1, long-term human-robot interaction,” in 2008 IEEE/RSJ Inter- pp. 64–73, 2011. national Conference on Intelligent Robots and Systems, pp. 3230–3235, Nice, France, September 2008. [7] R. Randell, J. Greenhalgh, J. Hindmarsh et al., “Integration of robotic surgery into routine practice and impacts on commu- [22] R. C. Arkin, M. Fujita, T. Takagi, and R. Hasegawa, “An etho- nication, collaboration, and decision making: a realist process logical and emotional basis for human-robot interaction,” evaluation protocol,” Implementation Science, vol. 9, no. 1, Robotics and Autonomous Systems, vol. 42, no. 3-4, pp. 191– p. 52, 2014. 201, 2003. [8] C. Ray, F. Mondada, and R. Siegwart, “What do people expect [23] B. Mutlu, T. Shiwa, T. Kanda, H. Ishiguro, and N. Hagita, from robots?,” in 2008 IEEE/RSJ International Conference on “Footing in human-robot conversations: how robots might Intelligent Robots and Systems, pp. 3816–3821, Nice, France, shape participant roles using gaze cues,” in Proceedings of the September 2008. 4th ACM/IEEE international conference on Human robot [9] S. Strohkorb and B. Scassellati, “Promoting collaboration with interaction - HRI '09, pp. 61–68, La Jolla, CA, USA, 2009. social robots,” in 2016 11th ACM/IEEE International Confer- [24] M. A. Salichs, I. P. Encinar, E. Salichs, Á. Castro-González, and ence on Human-Robot Interaction (HRI), pp. 639-640, Christ- M. Malfaz, “Study of scenarios and technical requirements of a church, New Zealand, March 2016. social assistive robot for alzheimers disease patients and their [10] G. Hoffman and C. Breazeal, “Robotic Partners Bodies and caregivers,” International Journal of Social Robotics, vol. 8, Minds: An Embodied Approach to Fluid Human-Robot Collab- no. 1, pp. 85–102, 2016. oration,” in AAAI Conference on Cognitive Robotics, 2006. [25] E. Law, V. Cai, Q. F. Liu et al., “A wizard-of-oz study of curiosity [11] M. Heerink, B. Vanderborght, J. Broekens, and J. Albó-Canals, in human-robot interaction,” in 2017 26th IEEE International “New friends: social robots in therapy and education,” Interna- Symposium on Robot and Human Interactive Communication tional Journal of Social Robotics, vol. 8, no. 4, pp. 443-444, (RO-MAN), pp. 607–614, Lisbon, Portugal, August 2017. [26] D. Hebesberger, T. Koertner, C. Gisinger, and J. Pripfl, “A [12] B. Robins, K. Dautenhahn, R. T. Boekhorst, and A. Billard, long-term autonomous robot at a care hospital: a mixed “Robotic assistants in therapy and education of children with methods study on social acceptance and experiences of staff autism: can a small humanoid robot help encourage social and older adults,” International Journal of Social Robotics, interaction skills?,” Universal Access in the Information Society, vol. 9, no. 3, pp. 417–429, 2017. vol. 4, no. 2, pp. 105–120, 2005. [27] J. Miehle, I. Bagci, W. Minker, and S. Ultes, A Social Compan- [13] M. N. Nicolescu and M. J. Mataric, “Learning and interacting ion and Conversational Partner for the Elderly, Springer Inter- national Publishing, Cham, 2019. in human-robot domains,” IEEE Transactions on Systems, man, and Cybernetics-part A: Systems and Humans, vol. 31, [28] T. Kanda, M. Shiomi, Z. Miyashita, H. Ishiguro, and N. Hagita, no. 5, pp. 419–430, 2001. “An affective guide robot in a shopping mall,” in Proceedings of [14] M. L. Traeger, S. Strohkorb Sebo, M. Jung, B. Scassellati, and the 4th ACM/IEEE international conference on Human robot N. A. Christakis, “Vulnerable robots positively shape human interaction - HRI '09, pp. 173–180, La Jolla, CA, USA, 2009. conversational dynamics in a human-robot team,” Proceedings [29] Y. Iwamura, M. Shiomi, T. Kanda, H. Ishiguro, and N. Hagita, of the National Academy of Sciences of the United States of “Do elderly people prefer a conversational humanoid as a America, vol. 117, no. 12, pp. 6370–6375, 2020. shopping assistant partner in supermarkets?,” in Proceedings [15] E. S. Short, E. C. Deng, D. Feil-Seifer, and M. J. Matarić, of the 6th international conference on Human-robot interaction “Understanding agency in interactions between ChildrenWith - HRI '11, pp. 449–457, EPFL, Lausanne, Switzerland, 2011. autism and socially assistive robots,” Journal of Human-Robot [30] H. P. C. Sirithunge, M. A. V. J. Muthugala, A. G. B. P. Jayase- Interaction, vol. 6, no. 3, pp. 21–47, 2017. kara, and D. P. Chandima, “Interpretation of interaction [16] T. Ozeki, T. Mouri, H. Sugiura, Y. Yano, and K. Miyosawa, demanding of a user based on nonverbal behavior in a domes- tic environment,” in 2017 IEEE International Conference on “Use of communication robots to converse with people suffer- ing from schizophrenia,” ROBOMECH Journal, vol. 7, no. 1, Fuzzy Systems (FUZZ-IEEE), pp. 1–8, Naples, Italy, July 2017. [31] L. D. Riek, “Wizard of oz studies in hri: a systematic review and [17] L. Tiberio, A. Cesta, and M. Olivetti Belardinelli, “Psychophys- new reporting guidelines,” Journal of Human-Robot Interac- iological methods to evaluate users response in human robot tion, vol. 1, no. 1, 2012. Applied Bionics and Biomechanics 13 [32] S. R. Klemmer, A. K. Sinha, J. Chen, J. A. Landay, N. Aboobaker, and A. Wang, “Suede: a wizard of oz prototyp- ing tool for speech user interfaces,” in Proceedings of the 13th annual ACM symposium on User interface software and tech- nology - UIST '00, pp. 1–10, San Diego, CA, USA, 2000. [33] C. L. Sidner, C. Lee, and N. Lesh, “The Role of Dialog in Human Robot Interaction,” in International Workshop on Language Understanding and Agents for Real World Interac- tion, Hokkaido University, Japan, 2003. [34] C. Sirithunge, A. B. P. Jayasekara, and D. P. Chandima, “Pro- active robots with the perception of nonverbal human behav- ior: A review,” IEEE Access, vol. 7, pp. 77308–77327, 2019. [35] I. Ajzen, “From intentions to actions: a theory of planned behavior,” in Action control, pp. 11–39, Springer, 1985. [36] B. H. Sheppard, J. Hartwick, and P. R. Warshaw, “The theory of reasoned action: a meta-analysis of past research with rec- ommendations for modifications and future research,” Journal of Consumer Research, vol. 15, no. 3, pp. 325–343, 1988. [37] I. Ajzen, “Perceived behavioral control, self-efficacy, locus of control, and the theory of planned Behavior1,” Journal of Applied Social Psychology, vol. 32, no. 4, pp. 665–683, 2002. [38] H. P. C. Sirithunge, M. A. V. J. Muthugala, A. G. B. P. Jayase- kara, and D. P. Chandima, “A wizard of oz study of human interest towards robot initiated human-robot interaction,” in 2018 27th IEEE International Symposium on Robot and Human Interactive Communication (RO-MAN), pp. 515–521, Nanjing, China, August 2018.

Journal

Applied Bionics and BiomechanicsHindawi Publishing Corporation

Published: Feb 23, 2021

References