2.1 Politeness
Most previous research on politeness in HRI has adopted Brown and Levinson’s [20] theory of politeness, which centers on the concept of “face.” The gist of the theory is the protection of ‘face’ or image by the social actors in a public domain. According to this theory [20], there are four strategies that a person could take to mitigate the “face-threatening” acts. The person can mitigate the situation by using an on or off-record strategy (on-record includes bald, positive and negative strategies; off-record strategy is to be indirect, using irony or metaphor). An actor can go on-record either without a redressive action (actions which are taken to minimize or overcome the intention of face-threatening) termed as bald strategy(being direct and clear in its strategy) or with a redressive actionwhich includes two different strategies, namely positive and negative. In positivestrategy, the face-threatening act is minimized byagreeing, being friendly, being optimistic etc.whereas in negative strategy,the face-threatening act is minimized by avoiding conflicts inshowing consideration.Based on this theory, a humanoid mobile robot was used to remind a user about medication while the user was busy with a primary task[17]. The studyinvolved four types of polite strategiesi.e., bald, positive, negative, and a mixed strategy (combination of positive and negative). Results revealed that negative and a mixture of positive and negative strategies were recommended for polite behaviors. The positive strategy in Brown-Levinson theory was discouraged. A series of study, which includes showing static pictures and animated clips to the participants, with a gatekeeper (peacekeeper) robot interacting with a human revealed that a polite strategy influenced the interaction[16].The participants noted that the robot with the polite behavior was friendlier, fairer, and acted appropriately. It also revealed that the polite robot was less threatening irrespective of the static picture and animated clip.Another robotic receptionist study [13][15] incorporated Brown and Levinson’s theory to develop a polite strategy with positive politeness. The studyapplied the bald strategy for the control group in two tasks:a chitchat task and a direction giving task. The polite strategy was positively perceived by the users in both tasks. The implemented polite behavior did not affect the HRI performance in the direction giving task. However, in the chitchat taskthe polite behavior with a humanoid robot impactedpositively the user perception. A study on compliance with a robot in relationship to speech and gesture features that express politeness suggested the need to develop multimodal levels of politeness since too much politeness caused negative impact [14]. The polite gestures, however, were positively associated with the social robot’s compliance. In a study in which adaptive feedback was implemented for a companion robot [18] the polite strategy was favored by the male participants while female participants preferred the direct commands. Another study examined the impact of impolite behavior on the performance of the participants in a physical trainer exercise [21]. The researchers found that the impolite robot (which was actually implemented as a rude robot) was not preferred by the participants. However, it yielded improved performance probably since it challenged the users.These studies support the hypothesis that polite behavior is preferred while interacting with a robot. However, all studies used humanoid robots.It is therefore crucial to expand the evaluation of politeness to other types of robots.
The concept of “face” and its implications for politeness rules may not be particularly suitable for human-machine interactions for several reasons. First, the strong emotional content associated with the face concept appears too strong for human-robot relations. Second, politeness based on face-saving strategies relies on verbal communication, whereas much of the interactions between humans and robots relyon nonverbal actions. Third, the face concept is highly sensitive to cultural variations[19][20]. Furthermore, it is inapplicable to many HRI tasks in which the robot does not include a face (such as industrial and other tasks). Finally, Brown and Levinson’s theory is relatively complex, and cannot be easily transformed into HRI design guidelines.
To circumvent some of these issues, Bar-Or et al.[24]proposed a theoretical framework for politeness in the field of HCI which was inspired by Lakoff’s theory of polite behavior[3]. In the context of HCI, they demonstrated that polite behavior has a positive impact on user perception and efficiency. However, it remains to be seen whether Lakoff’s work can be applied to the design of social robots and its effect on aspects of human-robot interaction.
Lakoff [3] suggests three rules for polite interaction: 1) Don’t impose your actions or views on other people (at least not without first asking for permission); 2) Give options to other people to let them make their own decisions; and 3) Be friendly while interacting with other people, in the sense of producing at a sense of equality between the parties. Compared to other prominent politeness theories [25][26], we consider Lakoff’s theory better suited for HRI research because it covers not only nuances of verbal interactions but also more general behavioral communication, which is an important aspect of social robotics. Unlike previous work on politeness in HRI, in which interaction was with a humanoid robot, the current study focuses on developing and evaluating polite behaviors in collaborative tasks with non-humanoid robots -- a robotic arm and a mobile robot.Further, we focus on politeness in the interaction itself (and not on polite robot behaviors related to motions and gestures such as approach distance, angle and speed).
To provide a comprehensive analysis of the influence of the robot’s polite behaviors, we investigated several parameters as detailed below. For this study, we conducted both live experiments and video-based experiments to try and isolate the effectof the moving robot and focus on the interaction aspects. The current study also includes a diverse population to explore the impact of politeness among different age groups, namely old and young adults. Lastly, we included two tasks (and related robots) to test the impact of polite behavior irrespective of the robot or task.
2.2 Study condition: remote vs. in situ.
The Covid-19 pandemic posed serious limitations on our ability to conduct ordinary HRI research. But as sometimes is the case, it also offered an opportunity to enrich the research scope and methods. Therefore, we conducted two types of experiments – one in video, during periods of strict social distancing, and one in situ, during periods of relaxation in social distancing measures. Beyond the practical constraints, the use of a remote (video) study was motivated by findings of a previous study [19], which pointed out that participants (old adults) were more focusedon the robot actions rather than concentrating on the interaction medium.However, as mentioned above, our goal was to assess people’s perceptions of the interaction rather than the robot’s physical movements. A video experiment was supposed to mitigate the saliency of physical activity and to help users focus on interactivity.The general guideline for conducting the remote experiment during the COVID-19 pandemic time has been demonstrated in [27]. Previous studies [28][29]suggested that video experiments could be used for exploratory studies in HRI. These studies evaluated the preferable approach direction for a robot in both video and live HRI trials and revealedcomparable people’s perceptions. However, both studies were limited to a university population which might have influenced the results (in [28],15 participants aged 21-56 with many of them with computer sciences or robotics background; in[29], 42 university students and staff, aged 18-56). These studies suggested that videos could be used for HRI exploratory studies. Simultaneously, however, they noted limitations – the more the interaction between the robot and the study participant in a trial the less suitable a video would be since it lacks important aspects of the interaction such as dynamics, embodiment, and contingency [29].Thus, we conducted experiments both with a video and a live experiment. In the video experiment, the users interacted with a robot that was remote from them. In the live experiment, the users interacted with a robot that performed the task in front of them.
2.3 Effect of age and gender on interaction
Effectiveness of assistive robots highly depends upon the acceptance and adoption by the users[30]. Far from the common perception that old adults are wary of technology [31], it was found they are open to new robotic technologies [32][33]. Attitudes regarding the robot, either regarding the social impact and comfort of the robot or negative towards the robot, were similar in case of old, middle-aged and younger adults[34]. The older old adults (75-84 years) found that a physical training robot (‘Gymmy’) was more useful as compared to their 65-74 year old counterparts[35]. The older group perceived the robot to be more useful compared to a 65-74 year-oldgroup. As aforementioned, politeness has been explored with different age groups [16]. However, the results did not reveal any significant effect of participants’ ageon the user perception of the robot.
Nevertheless, in the currentwork, we relate to the effect of participants’ age on user perception. In a person following feedback design study it was observed that perception, preferences and the attitude of users towards the robot highly depends on age and gender of the user [41]. A comparative study between a real and virtual humanoid robot [42] revealed that a greater number of old adults complied with the real robot and had positive impression of both robots but felt more attached to the virtual robot. A survey aimed to assess preferences of robot tasks among old and young adults found that old adults anticipate more benefits of monitoring-type robots [43]. Another study with old and young adults interacting with a humanoid robotin a cognitive training task revealed that the design of the robot and interaction should be adapted to the user’s age and needs[44].A previous preliminary study [19] in which we implemented polite behaviors for a robot manipulator revealed that young adults were able to differentiate between politeness levels. However, the old population was not able to do so. Both populations, though, indicated a preference for the polite behaving robot. All the above, barring one [16], suggested age is a relevant factor of interaction in HRI.
Previous research pointed out that gender plays an influential role in developing a perception about the behavior of the robot [16][18][35][36][37][38][39][40]. These studies suggested that male and female participants perceive the interaction with the robots differently. On the one hand,it has been suggested that male users are more aware of technological advancements than female counterparts [39]. Hence, male users tend to adapt to usage of robots more easily. This was also supported in [16], which was discussed in previous section, and included a comparison between male and female participants. The study found that the male participants perceived the polite robot more positively than female participants. On the other hand, [40]have found that female participants perceived the interaction with the robot more positively than male users, whereasin the polite gatekeeper the effect of gender on users perception was not significant [16]. Based on these ambiguous findings, we expectedthat gender would have no effect in our study.
2.4 Type of robot or task
HRI taxonomy is classified into three categories: interaction context (e.g., field of application, type of interaction), robot (e.g., task of the robot, morphology) and team classification (e.g., role of each agent, composition of the team) [45]. In Section 2.2, we discussed the interaction context,specifically with an experiment in video and liveconditions. In addition, this work involvesthe usage of different types of robot i.e., mobile and manipulator. Previous research mostly concentrated on comparing different anthropomorphic type robots [46, 47]. However, since the type of robot and task influences the interaction [48] it is important to consider this in the evaluation. Hence, we evaluated the effect of polite levels in two different robot tasks/types.
The task of both robots was to bring the utensils for table setting. However, there was a difference in the task typeperformed by the robot.In one case,the robot manipulates the utensils in the environment to achieve its goal (using the manipulator the robot brings the utensils in front of the user). In the other case, the robot transports the utensils from one place to another (the mobile robot transports the utensils from one room to another where the user is sitting).
2.5 Research questions
The present study aims to investigate the influence of polite robot behaviors on HRI evaluationswhen using different types of robots (stationary and mobile), in varying conditions (video and live), and in different age groups (old adults and young adults). The following questions were investigated in all four studies:
- Are participants able to perceive the differencesinpolite levels (irrespective of robot type, study condition, and age and gender)?
- Do participants prefer the polite behavior of the robot (irrespective of robot type, study condition, and age/gender)?
Though previous research revealed that differences in some measures depend on the type of robot or task, the overall level of automation had an effect on the interactions irrespective of the robot or task[48].Further research showed a similar impact of video and live experiments onhow users perceived the interactions with the robot [25][26].Additionally, previous research [16] points out that the polite behavior of the robot was preferred irrespective of the participants’ age.Using Lakoff’s politeness rules as a blueprint for the design of interactive robots, we believe that participants would be able to find the difference between the three different polite behaviorsand that they will prefer the politest behavior irrespective of their age, type of robot, and experimental condition.