Drawings for Insight on Preschoolers' Perception of Robots

The design of robots and robot-mediated activities for children needs to be informed by their expectations on robots, to ensure acceptability and effectiveness. Children drawings are a powerful tool to understand and describe these expectations, containing actionable insights for robot designers. We report a preliminary study of 50 drawings made by preschool children and investigate (i) their perception of robots and (ii) the change in perception induced by a short experience with real robots. Our analyses reveal that the children's age not only influences their perception of robots, but also how their perception changes after encountering robots.


Robot Demo
Figure 1: A subset of the children's drawings of robots ( =50) made before and after participating in a robot demo, clustered according to the age group.Arrows link the drawings that are likely to be drawn by the same child (since no identifable information was collected in the study, the correctness of the association cannot be guaranteed).

INTRODUCTION
As robotic technologies continue to permeate preschool children's daily life as toys, game companions or tutors, understanding children's perception of robots has become critical for Child-Robot Interaction (CRI) research [3,5,7,18,20].At the same time relatively little research exists on how preschool children interpret and understand robots [18], not in the last place due to the fact that traditional methods based on questionnaires and self-reports are difcult to implement at this age [4,19].Children drawings have long been identifed in child psychology as a "mirror to their mind" and a powerful tool to access the child's representation of the world [9,21].Indeed, various studies employ drawings to reveal and compare how preschoolers perceive robots before and after encountering them [5,8,10,20].However, few works use drawings outside of participatory design settings, no guidelines exist for analysing the drawings, and no conclusive fndings have been drawn concerning children's conceptualisation of robots.Several studies [8,10] incorporated drawing procedures primarily to engage children in interviews, rather than to analyse the drawings themselves.Others [5] viewed preschoolers as a small part of a much broader spectrum of participants (up to 13 years old) and gave limited importance to the role that drawing analysis could play towards understanding the conceptualisation of robots across diferent age groups.Finally, one study concentrated solely on preschoolers' drawings [20], aiming to identify diferences in robot depictions before and after two months of weekly interaction with educational robots designed to teach coding skills.As the study did not reveal any meaningful diferences, the question of insights into children's perception of robots derived from drawings remains under-explored and unanswered.
This paper is an efort towards the development of user-centered design for robots and CRI activities tailored for preschool children.Achieving this goal requires identifying information about children that can infuence acceptability and efectiveness of CRI, e.g., their perception of robots, developing tools to reveal it, and fnally using it to ofer guidelines that engineers can rely on.To this end, the research team spans a range of competencies, including child psychology, technology assessment, robotics and HRI 1 .
Concretely, we report a preliminary study involving 50 robot drawings made by preschool children before and after encountering real robots, that aims to investigate (i) their perception of robots and (ii) the change brought to this perception by a short experience with versatile collaborative robots.To analyse the drawings, we propose a scheme to characterise the elements in each drawing using a set of formal categories.Using this scheme, we describe signifcant diferences in how the 4-to 6-year-olds conceptualise the robot, and discuss several insights to the design of robots and robot-mediated activities in kindergartens.

METHODOLOGY 2.1 Study design
In this study2 , we collected two sets of robot drawings made by preschool children enrolled in a local kindergarten: the frst one within the week before they visited a robotics laboratory at the local university, and the second one within two weeks afterwards.For the drawing, the children were given a paper template, prompting them to imagine, draw and name their own robot.
The visit to the laboratory introduced the children to versatile collaborative humanoid robots, via a two-part demonstration of approximately one hour.The frst part involved the ARMAR-III robot [2] in a kitchen setting (Figure 2a) while the second part involved the ARMAR-6 robot [1] in an industrial setting (Figure 2b).In both parts, the robots performed various recognition and manipulation tasks in collaboration with a researcher.In the kitchen setting, the robot handed items from the table and the fridge, and took care of putting the dishes in the dishwasher.In the industrial setting, the robot collaboratively inserted a cover, as shown in Figure 2b.In both parts, the researcher ofered children simple explanations about the robots' abilities and components, engaged them via questions, and invited them to interact with the robots via object handovers.
We collected a total of 50 drawings.The frst set contains 31 drawings, made by 16 females and 13 males, aged 5.03 (0.89) years old. 3 The second set contains 19 drawings, by 11 females and 8 males, aged 5.21 (0.83) years old.Most, but not all, children participated in both drawing activities.With no identifable information collected, we are not able to provide the exact number of overlaps.

Coding scheme to analyse the drawings
To analyse the drawings, we developed a coding scheme with 12 categories (C1-C12) to describe each image using a uniform set of formal characteristics.To create the coding scheme, we initially  documented all the elements children used to depict a robot and identifed the frequently occurring elements that varied across individual images.The coding scheme was initially created by the child psychologist, and refned over interactions with two HRI researchers.The resulting scheme was then used by the child psychologist and two other researchers (not involved in its development) to independently rate the set of 50 drawings 4 , yielding three raters per drawing.The three raters then discussed the discrepancies and reached a consensus, resolving disagreements by majority voting.
The frst three categories in the coding scheme relate to the head, mouth and eye(s) shape (C1-C3).We observed that almost all robot drawings had a common basic structure, similar to how children of this age draw a human [11,16], including head, torso, arms, and legs (see Figure 1).In the depiction of the head, constant elements such as eyes and mouth were also present, schematically drawn as a human face.While this depiction reveals many similarities with a human fgure drawing, important diferences are present.For instance, eyes, traditionally round in human face drawings, in robot depictions could be both round (e.g., drawing D6 in Figure 1) and square (e.g., D12).Similar diferences are observed in the depiction of the head and mouth.Conversely, robots' torso and limbs are almost always similar to those elements in a human fgure: parallel straight lines and elongated rectangles.Being of little distinctive value, torso and limbs were not included in the coding scheme.On the other hand, we kept the more descriptive features (head, eyes and mouth depiction) for image analysis in order to determine to what extent the concept of a robot is linked to that of a human.
Additional elements, such as hair, dress, fngers, feet, buttons, antennas, ears, wings, etc., are present in the majority of the drawings.We classifed these elements based on their human, animal and mechanical nature (C4-C6), as they help determine to what extent the concept of a robot is linked to the image of a living or non-living being.From this list we excluded some elements which were not of comparative (diferential) value, namely feet, neck (as they were invariable in drawings) and fngers (as children at the age of 4-6 often draw fngers like set of "sticks", which makes it impossible to characterise them as human, animal or mechanical).
Variations in the level of image elaboration were also noted, from rather schematic and monochrome (e.g., D11), to having more details and diferent colours (e.g., D9).We thus introduced the categories image elaboration (high vs. low based on the number of details in the drawing) and colouring (whether the robot's contour is empty or coloured) (C7-C8).These categories aim to describe the child's engagement in the activity and interest in the subject (refecting the time and efort spent on the image creation).
Furthermore, we consider the amount of distinctive features (of animal, human and mechanical nature) included in the drawing (C9-C11).This metric, called features variety, is intended to disclose how the concept of a robot is refned with new experiences, by counting the various features attributed to robots.In contrast to the previous categories of colouring and elaboration, this measures how accurately the children understand what a robot is and is not.
Lastly, the names children gave to the robots varied from those similar to human names (e.g., Leo, Rosi) to those not typical for humans (e.g., A7, Herus).We capture this in the name type category C12.The choice of the name should reveal how distinctive in child's representation the robot is, compared to the familiar circle of human-like names owners (children, adults, pets, favourite toys, etc.).Not giving a name can indicate uncertainty in what name is applicable for the robot.

RESULTS AND DISCUSSION
A total of 43 drawings were considered for the analysis from the initial set of 50.We excluded 2 drawings due to unreported age, 2 for not following the task and 3 for lacking necessary features for analysis (e.g., recognisable head and torso).Figures 3 and 4 report the results for C1-C12 conditioned by age.
Robot images in the initial drawings of the 4-year-olds (see Figures 1a and 3a-left bars) closely mirror the familiar images of humans, living beings, fctional characters encountered in their daily surroundings, books or cartoons (as denoted by the prevalence of round heads, mouths and eyes, as well as the equal likelihood of human and animal features).These robots may take the form of an ordinary girl (drawing D7 in Figure 1), fairy (D1) or prince (D3), or an unknown creature (D2), amalgamating real features in the unique way children perceive them.Encounters with real robots do not seem to signifcantly alter the initial representation (see the paired drawings in Figure 1a), rather, they may contribute some distinctive and appealing features to the general image of a living being.For example, in D6, a child incorporates magic cameras capable of resizing the robot, as commented by the child.As a whole, we observe attempts to recreate previous images, often unfnished (D5 and D7) due to the lack of motivation (Figure 4a).
At the age of 5 (see Figure 3b), the initial drawings are mainly based on the typical human image (including round eyes, mouth, eyelashes, hair, dress, etc., in D9-D11), partially complemented by animal (horn in D10) or non-human-like features (squared head or mouth in D8, D10, D11).Children at this age may still lean towards the imaginative world of the 4-year-olds, but they also probe the more reality-oriented approach of the 6-year-olds.Exposure to real robots at this age induces substantial changes in the robots representation: human features become less prominent and mechanical features increase notably.Squared non-human-like elements replace rounded forms on the "face" (pair D10-D12).5-year-olds experiment with replicating observed robot appearances (D14), substituting point-like human eyes with large round camera-like features, or changing head shape to half-round and incorporating wheels instead of limbs (pair D11-D13).
The initial drawings of 6-year-old children show more stereotypical characteristics of a robot (see Figure 3c), aligning with how older children depict robots [6,13,17].These drawings, beside squared parts of the robot fgure, often feature common attributes such as antennas, grippers, buttons, spring-like arms (D15 and D18) or signs of superhero abilities (D16).At this age, children shift their focus towards reality, becoming more attuned to the social world around them, its rules, norms, shared knowledge, etc. [12,15].Exposure to real robots imparts new information, compelling children to integrate this knowledge into their drawings.Post-demo robot images show a reduction in stereotypical features, such as square forms for heads, eyes, mouths (robots may lack the mouth entirely D22) and features like antennas and gripper-like hands (D20-D22).Children, drawing inspiration from their new experiences, try to depict more accurately how real robots are designed (D20), emphasise their functionality (e.g., placing it in the kitchen) and operation (e.g., cables leading to sockets in D21).
We found that the initial interest in drawing robots, estimated through image elaboration and colouring, was rather high (Figures 4a and 4b), even considering that the novelty of the task might have contributed to it.Revisiting the same task after encountering a real robot appeared less interesting, especially for the 4-yearolds.Some children simply re-made their frst drawing (D2-D5 and D17-D19).Figure 4c shows that image elaboration in post-demo drawings decreased mainly by removing additional human-like features from robot images, reducing their diversity to one element in each age group (prior to exposure, their number ranged up to 6 elements).At the same time, the variety of additional mechanical features increased in all age groups, most notably at the age of 5 (Figure 4e).These changes not only indicate that real experience contributed to clarifying the robot concept (making it more distant from human-like images) but also show that some children sought to enrich the image of the robot with more specifc and diverse features, potentially showing higher interest in gaining knowledge about robots.The same tendency is revealed in robot names: the children tended to give specifc (non-human-like) robot names in post-demo drawings, compared to pre-demo (Figure 4f).

CONCLUSION
Albeit preliminary, the results suggest a correlation between the age of the children and the features of robots in their drawings, allowing us to characterise the initial robot perception in diferent age groups, as well as how a brief but real experience with robots can infuence the initial concept.Based on our comparative description, we can tentatively outline some guidelines for design elements that seem to be important for diferent preschool ages.For the 4-to 5-year-olds, the robot's appearance should not be tied to a defnite image, leaving room for invention and playing out their scenarios (as suggested by the variety of feature types and inspirations that characterize the drawings of this group).At the same time, the robot's expressiveness seems to be crucial: for example, [14] observed that children aged 3-4 displayed their most active reactions when the robot's eyes changed into a heart shape and cheeks turned red.Conversely, the marked changes in the type of features used in the post-demo drawings suggest that for 5-to 6-year-olds, understanding the robot's functionality could become more important, as they might like to discover how to control and manage the robot's actions.This assumption relates to the study in [18], where children of 4-to 7-year-olds answered a question about robots' purpose ("what are robots for?").
These conclusions are indeed preliminary.We hope to fnd further evidence from the ongoing feld study in the kindergarten with two social robots interacting with children in diferent activities.At the end of this study, along with the data about children's engagement, we will analyse the third set of robot drawings, which will hopefully give us additional insight into the robot concept and whether it changes after a prolonged social interaction with a robot.

Figure 2 :
Figure 2: Robot demonstrations shown to the children.