Date created: | Last Updated:

Category: Project

Description: The use of social robots, defined here as physically embodied artificial agents that humans can interact with in-situ, has become increasingly common over the past decade (Johal, 2020). Social robots have been deployed across a variety of learning tasks (affective and / or cognitive), in a variety of roles (48% as tutor, 38% as teacher, 9% as a peer or novice co-learner), and in various educational and research settings (Belpaeme et al., 2018). Compared to virtual agents or video/voice-based tutors, social robots offer promise for educational settings where it is well documented that social and physical interaction supports human learning (Belpaeme et al., 2018). In a comparative review, more positive outcomes were associated with tasks where the artificial agent was both physically embodied and co-located, compared to tele-present robots or other virtual agents (Li, 2015). Overall, social robot research in education provides evidence of positive affective and/or cognitive learning outcomes compared to other artificial agents (Belpaeme et al., 2018). For instance, social robots have been shown to improve cognitive gains in a puzzle task, compared to virtual or voice-only tutors (Leyzberg, et al., 2012), and increase task compliance, enjoyment, and performance in other tasks (Belpaeme et al., 2018). However, what it is about the physical presence of the robot that drives these performance gains remains not yet well understood. By convenience, much of the research with social robots in educational settings uses existing and readily available robots. Nao, a relatively inexpensive humanoid robot is the most popular option, used in 48% of studies reviewed by Belpaeme and colleagues (2018). Beyond this, however, the range of social robots used in educational research settings is highly varied across the robot’s form, features and functional capabilities, and few studies have directly compared the effectiveness of different types of social robots across equivalent tasks (Papakostas et al., 2021; Hortensius & Cross, 2018; Cross & Ramsey, 2021). The form and function of these off-the-shelf robots may influence performance outcomes in ways that are not yet well understood (Henschel et al., 2021). Social neuroscience research provides powerful evidence that human--robot interactions (HRI) are influenced by the beliefs, perceptions and expectations users hold about the artificial agents they engage with (e.g., Hortensius & Cross, 2018). Many different factors may influence how we perceive a robot, and what we feel or how we act in their presence (Cross & Ramsey, 2021), including what the robot looks like, and its perceived capabilities. For example, Woods (2006), in exploring the design space for robot development, provided evidence that it is the overall appearance, rather than just the facial features, that influence a child’s perception of the robot. Children were consistent in relating different forms in design, including shape, anthropomorphic features, and modes of movement, to differing personality characteristics. For example, aggressive robots were seen to be more machine- like, and happy / friendly robots more animal-like (Woods, 2006). In support of this, Hwang et al (2013) also found that the shape of the robot influences the emotions and personality attributed to the robot. Despite this, there is limited research into how the appearance and functions of a robot influence perception and expectations in educational tasks. It will be important to build a clear understanding of these issues if robots are to be used effectively to support educational outcomes in the future. To begin to address this, Caruana, Moffat, Blanco & Cross (in prep -https://osf.io/jdv2y/?view_only=f0e43b6b292747d587468865c2a02dd2), sought to compare the first impressions, expectations, and subsequent experiences amongst early learners during a reading task with three different commercially available social robots (Nao, MiRo, Cozmo) that varied across both form and function. A key finding from this study supports the view that the aesthetic and functional capabilities of the robot influences perceptions around intelligence and friendliness (Caruana, Moffat et al., in prep). Nonetheless, a limitation of this recent study remains the pragmatic necessity to use available robots, which constrains our ability to explore what children want and expect from learning-assistant robots before they set foot in the lab (or classroom). Why reading? Different educational tasks (and learner needs) influence what is required and expected from a social robot from the perspective of the user. Exploring form and function within a specific task keeps the task variable constant. Reading is a universal learning task – all children need to learn to read – and require frequent practice and feedback on their progress in order to do so. It is also one that can be difficult for some children, who require more intensive teaching, involvement, or other interventions. Approximately 16% of children struggle to learn to read for a variety of reasons (Shaywitz et al., 1992), and difficulty with learning to read is associated with various internalising conditions, including anxiety, depression (Francis et al.,2021) and reduced self-concept (Mcarthur et al, 2020). Whilst this can be addressed with combined and targeted reading instruction and psychological interventions (e.g., cognitive behavioural therapy) (Francis et al., 2021), tools which support children’s self-engagement in reading outside one-on-one intervention settings are also needed. Emerging evidence from reading-assistant dog programs suggest that the presence of a non-human and non-judgemental social agent can support reading outcomes in children (see Hall, Gee & Mills, 2018 for a systematic review). Along these lines, social robots may offer a more practical, accessible, and flexible alternative to dogs, as they have the capacity to offer a non-judgmental social presence, but do not require the same sort of care and expert handling as a live animal does and can potentially offer more forms of relevant feedback (Caruana, Moffat et al., in prep). A design-based approach Ample research exists exploring the necessary computational elements that create a “social” robot – such as responding appropriately to cues, understanding social rules, maintaining eye contact as appropriate, and animacy in voice intonations and gestures (Breazeal, 2003). However, setting aside the complex task of programming social robots to be appropriately responsive to the cues and adaptive to individual needs of the moment, a simple question has yet to be asked of the children who are expected to engage with these robots, without biasing their perceptions: What is it young children want or expect in a reading buddy robot? Current Study. A child’s perspective on what is desirable in a robot-assistant can and does differ from that of an adult (Lin et al. 2014). To this end, our current project sets out to explore which features matter to children in a robot designed to accompany them while they learn to read, free from the bias that may come from being presented with a physical robot in the experiment. Some work has been done to understand the design space for children’s robots previously. Woods (2006) found evidence that children prefer cartoon or toy-like characters with human characteristics, over more realistic robots, for example, with robots that are too human-like creating a sense of discomfort (the so called ‘Zombie’ effect, (Woods, 2006)). Lin and colleagues (2013) further explored expectations and design features for a service robot for children’s libraries. They also noted a preference for a more cartoon-oriented design, function-orientated movement, and visual feedback amongst children (Lin et al., 2013). However, this is the first time, as far as we can tell, young learner readers will be consulted as to their expectations and desires for a reading buddy robot. Unlike previous studies which have used pre-determined robot stimuli to probe the design space of robots (Wood, 2006), our aim is to use qualitative research via a “blank-sheet” discussion with young children of learning to read age (5-9 years) to build a better understanding of what they would like in a reading buddy robot. In a recent paper, Cross & Ramsey (2021) proposed a feature-based, or dimensional, framework to explore how robot variability influences human-robot interaction. Whilst not prescriptive in the dimensions proposed, they included size, human-like form or motion, functional repertoire, automated features, socialness, intelligence, and pre-existing expectations. Our open discussion approach will aid us in identifying and exploring which dimensions are spontaneously relevant from the child’s perspective, in a reading-buddy robot. Beyond this, we intend to probe for specific dimensions identified within Cross & Ramsey’s (2021) proposed framework, including physical form (shape, anthropomorphism, size, colour, material), functional capabilities (mobility, communication, feedback, actions), and character attributes (such as intelligence, friendliness, socialness).