"It's Not a Replacement:" Enabling Parent-Robot Collaboration to Support In-Home Learning Experiences of Young Children

Social robots represent a growing technology that could provide social interactions and physically embodied presence. These affordances have the potential to convert an individual learning experience into a socially-enriched one [41]. Compared to virtual agents, physically embodied robots can similarly offer adaptive assistance [34, 49, 50] and verbal communications [5], yet distinctly support learners’ engagement with the physical world and elicit higher-level social behaviors during learning [2], thereby promote better learning outcomes [35]. Prior work has shown that social robots can effectively support school-based learning in mathematics [36], science [11], and literacy [21, 32]. Similarly, they can enhance children’s learning during home-based activities such as reading [41, 42], number board games [26], and math-oriented conversations with parents [27].

Prior work has demonstrated how social robots could support children’s learning within varying social dynamics and stakeholders. We can categorize these studies based on the number and type of agents that are involved into interactions involving (1) a robot and a single child, (2) a robot and multiple children, (3) a robot, multiple children, and an adult, (4) a robot, a child, and an adult. First, a substantial body of work has focused on how a social robot could support a single child. For example, Kennedy et al. [32] demonstrated how children learn elements of a second language from the robot in a short-term dyadic interaction context; on the other hand, Tanaka and Kimura [55] showed how a child can take the role of a teacher to instruct the robot as a novice, gaining self-confidence and improving learning outcomes. Chen et al. [9] suggested an adaptive role-switching model to flexibly accommodate a child’s personalized needs for being a tutor or a tutee dynamically in a one-on-one interaction with a social robot during vocabulary learning. In addition, beyond the context of structured learning, Ali et al. [1] demonstrated how an autonomous social robot can foster a child’s creativity using verbal and non-verbal behaviors.

Secondly, some studies have addressed how a social robot interacts with multiple children in educational contexts. For instance, Escobar-Planas et al. [18] investigated how children interact with a social robot in pairs during a problem solving task and found individual differences in the perceived role of the robot based on varying degrees of robot expressivity, suggesting the importance of considering individual perception when designing the robot to support children’s learning in groups. In addition, Kanda et al. [31] uncovered how social behavior exhibited by a robot supported the initial building of relationships among children for collaborative learning in a classroom setting.

Adding an adult to each of the dynamics mentioned above results in the third and the fourth forms of social dynamics with a social robot. The third form of interaction with a robot involves multiple children and an adult who could be a teacher using the robot in a classroom setting [61]. For instance, You et al. [61] introduced a humanoid robot into a English class as a teaching assistant to support the teacher motivating a group of children. The fourth form of interaction with a robot involves a child and an adult who could be a parent [6, 7, 8, 22]. Although several studies describe parental perspectives on incorporating a social robot in their child’s learning, most of them have not integrated a situated interaction experience with a social robot into their methods, relying instead on surveys or interviews facilitated by pictures of social robots. These studies suggested that parents hold a positive attitude towards the potential of integrating a social robot in their children’s learning either at home or school [33, 37, 45, 56]. However, they also reported several concerns including technical challenges [56], undesirable replacement of the teacher’s role [33], and disruption of children’s school-based learning [37]. A few studies we found incorporated a physical robot in study sessions with a parent and a child to induce insight into this interaction dynamic [7, 8, 22]. Gvirsman et al. [22] developed a robotic system called Patricc that promoted parent-toddler-robot (toddlers aged 1–3 years old) triadic interaction more than a tablet in a 8-minute learning session, measured by video-coded gaze. Another study designed a technology-based intervention, called WAKEY, for facilitating morning routines of preschoolers (aged 3–6 years old). In a week-long evaluation study, WAKEY was found to mitigate parental frustration about communication regarding morning routines with children and make positive impact on parents’ attitude in parenting and communication [7]. In addition, Chen et al. [8] designed a platform that integrated a Jibo robot and an Android tablet to facilitate parent-child co-reading activities in a 3–6 week long deployment study, targeting families with children aged 3–7 years old. They uncovered that social robots could improve various aspects of parent-child interaction, recommending that the robot’s role in a group setting should be explored from individual group members’ perspective and that these opinions should be further integrated in the design of the robot.

Parents are one of the most crucial socialization agents that influence the acquisition and development of a broad spectrum of cognitive and social skills in children [4, 14, 46]. Extensive literature in the field of educational psychology has consistently demonstrated the positive impact of parental engagement in children’s learning on their academic achievements [20, 25]. Moreover, active parental involvement within the home environment holds particular promise for enhancing learning outcomes, surpassing the effects observed in a school setting [19, 23, 54]. Furthermore, Hoffner and Buchanan [28] identified a correlation between the extent of parental guidance and children’s age, with younger children eliciting more parental mediation compared to their older counterparts. These findings underscore the importance of promoting parental involvement, especially within the home, in the education of young children.

As various forms of technology and media become increasingly prevalent, parents are increasingly taking an active role in managing their children’s use of these technologies. Parental mediation is critical for young children, as it further determines how children’s learning and development is induced by the media they are interacting with. Warren [60] defined parental mediation as “strategies parents use to control, supervise or interpret media content for children.” Parents mediate children’s use of technology using different strategies depending on various factors such as their confidence in using technology [12].

Parental mediation theory has been continuously expanded by the Human-Computer Interaction (HCI) community to explain the involvement of parents in children’s technology and media usage in a broader context. The foundational version of parental mediation theory originated from the patterns of watching television, encompassing three primary strategies: restrictive mediation, active mediation, and co-use [57]. Restrictive mediation refers to how parents make rules to guide or prohibit the viewing of certain content; active mediation (sometimes called “instructive mediation” or “evaluative mediation”) refers to how parents discuss the content with children during or after viewing; co-viewing refers to situations when parents and children watch television together, promoting parent-child connections [57].

Apart from television, parent mediation theory has also been extended to explain the use of emerging technologies such as the internet [15], video games [44], mobile devices/games [53], creation-oriented educational media [62], and conversational agents [3]. First, Nikken and Jansz [44] suggested that parental mediation strategies in video-game play for children are comparable to those for television-watching, after conducting a survey-based study. They further explained how specific contextual factors such as parents’ game behavior relate to these strategies. Moreover, parental mediation strategies were re-interpreted from traditional views for a location-based mobile game. Through a mixed-methods study, Sobel et al. [53] found that the location-based mobile game, Pokemon Go mitigated parents’ traditional screen-time concerns, as it encouraged outdoor activity and promoted opportunities for parent-child bonding. This study exemplifies how the use of a different kind of technology can shift the interpretation of a restrictive mediation strategy (e.g., restriction of screen time) from a traditional perspective (e.g., great concerns about screen time) towards a modified view (e.g., mitigated concerns about screen time). Furthermore, Beneteau et al. [3] recommended including the concept of augmented parenting in addition to the existing mediation strategies for a conversational agent. They investigated how voice-based interfaces, such as Alexa, affected parent-child dynamics at home and found that voice interfaces fostered communication between parents and children, disrupted each others’ access to the technology through interruption or parental control, and augmented parenting practices. They argued that as the access to the technology is democratized (i.e., all family members have independent access to the device), parents might use technology to both mediate and augment parenting practices. Finally, Yu et al. [62] created a different mediation framework for the context of using creation-oriented educational media, inspired by the parent mediation theory. The newly-inspired mediation framework included creative mediation, preparation mediation, and administrative mediation. Each of these components partially reflected the notion of the three strategies from the original framework (i.e., restrictive mediation, active mediation, and co-viewing), but afforded its own meaning tailored to the specific educational context.

The evolving framework or interpretation of parental mediation theory for different kinds of technology reflects the distinct needs of parental involvement, influenced not only by contextual factors but also by the affordances of the technology. However, very little has been done to explain the parental involvement needs for social robots in the context of education. In our work, we intend to explore how parents prefer and envision the adoption of a companion robot in their interactions with their child during in-home learning.

Overall, studies related to interactions between a robot and a child or multiple children are relatively abundant. However, more attention needs to be oriented towards including adults in these interactions, given the importance of mediating technology use, especially for young children. Although some work has been done to understand the interaction dynamic among a robot, a child, and a parent, questions related to how a parent would adopt a social robot to mediate the use of the robot and foster learning in children as well as how the design for such interaction should be like are yet unclear. We investigated a different age group from Gvirsman et al. [22] and a different learning activity from Chen et al. [8], while Chan et al. [7] is not relevant to a learning context but daily routines. To fill in the research gap, we aim to address the following research questions in this work:

After obtaining institutional research ethics approval, we recruited families from cities in the Midwestern United States via email distributed to community centers and university employee mailing lists. We selected participants based on the following criteria: (1) one parent and one child were required to participate in the study together, (2) the child must have been between 3 and 5 years old, and (3) both the parent and the child were able to understand and communicate in English. We selected this specific age group as it corresponds to a period when informal learning experiences with parents at home are relatively important [47], occurring before children’s active participation in formal schooling. Each family received $25 USD as compensation upon completion of the study.

Our analysis includes data from 10 parent-child dyads, each with a child aged between 3 and 5 years old (4 female, 6 male, M = 4.3 years old, SD = 0.58 years old). Each study visit involved one parent and one child. Other parents and siblings were sometimes present nearby but did not take part in the study activities. Participant demographics are listed in Table 1. We note that all parents who chose to participate in the study were female as mothers typically serve as primary caregivers [51] and are accordingly the ones who were willing to participate in the study. In addition, despite our efforts to recruit a diverse sample, our final sample offers limited socio-demographic representation (80% of the mothers have at least a Bachelor’s degree, 80% of families have at least $75,000 annual income), which may further constrain the generalizability of our results and omit insights into the unique needs of demographic groups that are less represented in our sample. We will discuss this limitation more thoroughly in Section 5.3.

We conducted short-term in-home visits with families and used a learning companion robot as a technology probe to foster discussion on how a parent would engage and use the robot to support their child in learning and how they envisioned the robot should be designed for parent-child interaction.

Our study objectives align with the three main goals described by Hutchinson et al. [30] about the technology probe approach–understanding users’ needs and perception about the technology in a real-world setting, field-testing a technology, and motivating design of new technology. With the limited understanding of how parents might adopt a learning companion robot to interact with their child, introducing the learning companion robot with simple and flexible functionalities allowed us to explore unknown needs and preferences of families, understand their interpretation of how to use the technology, and inform new design opportunities for future improvements. In the following sections, we describe the details of the study materials and procedure.

To anonymize our participants in the data, we report the findings using participants’ ID assigned by the experimenters based on their order of completing the study. For example, P1 refers to the first participant. We conducted a reflexive Thematic Analysis (TA) based on Clarke and Braun [10] and McDonald et al. [39] on the semi-structured interviews to identify patterns in the data and derive a grounded understanding of parents’ perspectives. First, the first author and an undergraduate assistant, who is experienced in qualitative coding, transcribed and familiarized themselves with the data based on the audio and video recordings. Second, the first author generated initial semantic codes based on her interpretation of the data and categorized the data based on the codes. For example, code “leverage robot’s responses to engage children” and code “use robot’s answers as a source of validation” were both categorized as “parents’ use of robot’s response.” Third, both the first author and the undergraduate assistant coded the data from P1 with the initial codes, followed by an iterative discussion to find consensus on interpretations and emergent candidate themes and to generate an agreed code book. Fourth, for the rest of the data, the first author and the undergraduate assistant used the code book to code the rest of the data independently and peer-reviewed each others’ codes. For instance, the first author coded P2 and the assistant peer-reviewed P2; the assistant coded P3 and the first author peer-reviewed P3. The finalized codes were reviewed and discussed iteratively. Lastly, we reviewed, refined, and constructed the themes as findings based on meaningful and insightful patterns.

In our data, parents frequently discussed several tasks (e.g., generating creative educational prompts for specific learning concepts, offering personalized scaffolding, etc.) that must be completed to achieve the learning goals they expected for their child. More specifically, they illustrated how these tasks should be delegated between themselves and the robot as they reflected on their as well as the robot’s strengths and weaknesses in daily learning scenarios. This pattern in our data suggested that parents were thinking more about how they could collaborate with the robot, which was further verified by their explicit description for the robot’s role as their “collaborator.” In addition, task delegation choices were based on three dimensions: parent/robot relative capability, parent availability, and parent motivation. A summary of these findings can be found in Figure 6.

Parents formulated their individual policies for using the social robot, while upholding their core educational objectives when engaging in learning activities with their child. Even when guided by prompts from the social robot, they instinctively tailored their parental guidance to fill in any gaps and ensure the attainment of these objectives. This not only reveals the parents’ aspirations in supporting their children but also highlights potential design implications for the social robot. We categorized these learning objectives into three domains: intellectual development, social and emotion development, and promotion of independence. A summary of these findings can be found in Figure 7.

The concept of interaction paradigm defines a specific module of interaction with the robot, governed by the interaction dynamic and the preferred initiation of communication. In our data, we found that parents preferred the robot to flexibly supplement their level of involvement and they favored taking more control over when the robot initiates an interaction. A summary of these findings can be found in Figure 8.

Parents wanted the robot to adapt to flexible choices of interaction dynamics.. The concept of interaction dynamics refers to how the robot manages the number and kinds of participants as well as the involvement of each party. Instead of adhering to a fixed mode, most parents (P4, P5, P6, P8, P9, P10) wanted the robot to flexibly support a range of interaction dynamics based on situational needs. For instance, P5 illustrated that she would like to “have choice to switch between having parents in the interaction or letting the child be on his own.” Parents conveyed a variety of dynamics they preferred in different scenarios with the robot, including child-robot interaction without a parent, parent-child-robot interaction, and parent-child interaction without the robot.

Firstly, in the dynamic of Child-robot interaction (without a parent), parents leaned towards allowing their child to interact with the robot independently when they are temporarily unavailable (P4, P5, P6, P8, P9), when they lack motivation (P1, P9), as the child matures to an older age (P2, P5), or when the objective is to nurture their child’s independence for self-directed learning (P5, P8). For instance, P5 described that when she was “trying to do three or four different things at the same time, [child] will be on his own.” Moreover, P9 indicated that she prefer to do more outdoor activities “rather than learning activities at home with her child,” and thus would favor a child-robot interaction model for in-home learning activities. In addition, P2 and P5 both imagined their children gradually interacting with the robot on their own as they get older. Lastly, P8 illustrated her desire to promote independent learning for her child, stating that the robot “provide the response that gives her the tools to be a little bit independent and practicing the skills without me” (P8).

Secondly, in the dynamic of Parent-child-robot interaction, all parents (P1—P10) expressed their preference for a flexible division of roles in guiding interactions with the child, allowing either the robot or themselves to take the lead while the other played a supplementary role. This adaptable approach results in varying levels of involvement for each party and thereby variations within this specific dynamic. Firstly, parents leaned towards increased robot involvement when they lack the availability (P5, P6), motivation (P4), or capability (P2, P8) to actively guide the activities, when they aim to cultivate their child’s independence in learning (P5). For example, P8 described the lack of experience her husband has in facilitating children’s learning, stating that “my husband does not know how to facilitate and he wouldn’t enjoy getting emails about it. The robot could guide his interaction and maybe have it model him[husband] more.” Conversely, parents themselves wanted more active involvement when they seek quality bonding time with their child during learning (P4, P6), when they intend to contribute specific intellectual concepts (P4, P10), when personalized support is deemed necessary, or when they want to maintain supervisory control over their child’s interaction with the robot to address any concerns (P7). P10 illustrated this when she stated “I think it is always beneficial from having some parental interaction and interest in what he’s doing. So I want to be there to encourage him and let him validate that he’s doing the job.” Combining these two views, parents prefer to have choice in both. P5 explained that she “would like to have choice” in the dynamics, and be able to “switch between involving parents to interact more and less.” In particular, she stated that “I can provide guidance or assistance, but only being pulled in when needed, letting him do what he needs to do by himself.”

Finally, in the dynamic of Parent-child interaction (without the robot), a few parents preferred to interact with their child without the robot when they want to enjoy their own time together (P9), when the concepts they want to explain are not enabled by the robot (P8), or when they have concerns over the information the robot offers (P6). For instance, P9 illustrated the situation when she said “I feel it is valuable to have one on one time with the child, doing reading or an activity together.”

Parents envisioned the robot accommodating flexible forms of interaction initiatives.. The notion of interaction initiative refers to how much autonomy the robot is granted to initiate interactions with users. Our data suggested different patterns of initiating an interaction including user-driven, user-authored and robot-driven, robot-driven with user permission, and robot-driven. In the following sections, we articulate each kind of initiative and the the possible scenarios correspondingly.

First, in a user-driven case, parents (P2, P3, P7, P8, P10) preferred the robot to initiate an interaction only when they manually request one. P3 described that she preferred the robot “to be manual, rather than talks on its own after a certain period of time.” Second, in a user-authored and robot-driven case, where parents prefer to author the timing and initiatives of the robot. P4 described that she “wouldn’t want the robot to just be able to barge in to say whatever it wants or ask for permission randomly.” She further illustrated her intention to author the robot’s initiatives by scheduling a specific prompt or behavior for a specific time frame, stating, “it[robot] can initiate to say, ‘it is time for this thing, do you want to do it now,’ and I can confirm if I want to do it now or later.” Third, in the case of robot-driven with user permission, parents (P2, P7) described certain scenarios when the robot can drive the initiative by indicating its intention but need to wait for user permission before actually putting the initiative into actions. P7 described that she preferred the robot to “raise her hand while we read together before bedtime” to keep the interaction “at a lower intensity level.” Finally, in a robot-driven case, parents (P6, P7) described that sometimes the robot can initiate an interaction or conversation anytime. P7 exemplified this case by describing a scenario when her child is playing toys: “There could be a time when she can just interact and interrupts. I can imagine her roaming around and say ‘let’s build a racetrack’ or something similar” to help come up with creative learning opportunities.”

We see an opportunity to extend our understanding of parent-robot collaboration and the factors that might affect the emergence and outcomes of this collaboration by connecting findings across the three themes emerged in our analysis, which we discuss below.

Here, we present the design considerations for the notion of parent-robot collaboration. We note that this idea extends beyond interactions actually involving a parent, a child, and a robot. Instead, we adopt a holistic perspective and include scenarios where the robot helps implement the parenting practices that parents would have done by themselves but were unable to carry out practically. In the upcoming sections, we articulate the meaning of each design implication and outline potential design considerations for practical implementation. Based on the discovery of three major themes and their connections, we have identified the following design implications:

A key limitation of our work is that our participant sample offers limited representation in the following socio-demographic dimensions: (1) location: all participants lived in the Midwestern United States; (2) gender: all parents were females; (3) income: more than half of the families have middle to high level of household income; and (4) education: more than half of the mothers had a Master’s degree or above. First, the parents self-selected which parent would participate, rather than the study team, and thus our entire sample was composed of mothers. This is consistent with prior literature in education [38, 51]. Despite our efforts to recruit a representative sample, it was challenging to balance multiple demographic factors (e.g., gender, race, income, maternal education, etc.) all at the same time. Plus, fathers at times signed up for the study but mothers ended up being the ones who participated. Similarly, due to common challenges of recruiting lower-income families [43], we have little low-income families in our sample. It is well-established that children in families with higher income or education levels tend to have better academic abilities [52], which may influence the kinds of support those families prefer the robot to offer. There may be specific needs and support that fathers or lower-income families prefer but were not well captured in our results because of lacking enough representatives in our sample. We will expand our efforts to recruit a more representative population in our future work.

We conducted brief home visits to gain insight into parents’ perspectives on using a learning companion robot. However, this short-term exploration may restrict our findings to the short-term effects and immediate parental experiences with the robot. It may not encompass the evolution of interactions over time, necessary updates, and design considerations for adapting to changes within an evolving ecosystem. For example, parents have voiced concerns regarding content appropriateness and the need for parental control. These sentiments may evolve over an extended period as parents develop stronger bonds with the robot and the robot demonstrates its reliability. Future research should employ long-term deployment approaches to examine how parents engage with the robot over an extended duration.

Finally, our work has presented qualitative findings concerning user perceptions, preferences, and design implications. To practically implement our design recommendations and better assess their feasibility, it is essential to construct viable prototypes and analyze the statistical relationships between the identified factors. Future research should prioritize the development of prototypes based on the design implications, evaluate the practicality of these design concepts, and quantitatively investigate the relationships between the factors.