Perception of Robot Personality Traits Based on its Design and Behavior
Study on QTrobot in Educational Human-Robot Interaction with University Students

Social robots are interactive physical agents capable of communicating, learning, and acting socially [12]. In education, they serve as teaching assistants, co-learners, and companions [1]. The human-like appearance and interaction capacity of robots can benefit students by helping them to develop critical thinking, solve complex problems, and understand difficult concepts through embodied learning activities [1].

Despite such advantages, several limitations remain. Most robots struggle with accurate speech recognition and interpreting complex social cues, thereby affecting classroom performance negatively [13]. Issues such as the robot's failure to connect with children at the beginning, unclear instructions, and unreliable behavior contributed to interaction breakdowns in long-term classroom studies [14], highlighting the need for more robust and natural interactions.

Robot personality is central to meaningful HRI [15]. UX can be enhanced by making robot behaviors predictable, intuitive, and enjoyable [7, 15, 16]. People naturally attribute personality traits to robots, using these traits to interpret and anticipate robot actions [17]. The “Five-Factor Model” (FMM) is the most widely used framework for developing and evaluating social robot personalities [1, 6, 8, 16, 18, 19], categorizing personality traits into five core dimensions: neuroticism, extraversion, openness, agreeableness, and conscientiousness [20]. Researchers map these traits to observable robot behaviors, enabling distinct personality profiles. For instance, researchers developed emotional expressions of a humanoid head robot utilizing various levels of extraversion and agreeableness [16].

Robot personality traits are the visible aspects of their personality [6], and users’ perception of personality traits can significantly influence overall UX. Perceiving a robot's personality significantly depends on its verbal and non-verbal communication [6, 7, 9, 10, 21]. Users infer traits like extraversion or introversion from appearance, voice, gestures, and expressions [7]. Consistent expression of these traits across modalities improves recognition and enjoyment [7, 10]. However, findings are mixed on whether matching or complementary personalities are more effective. Some studies reported higher engagement with complementary traits [7, 10], while others show improved performance with matching traits [15], suggesting that interaction goals and individual differences may moderate this effect. Appearance further shapes personality perception. Human-like characteristics can enhance perception of traits and relatability [3, 9]. However, when robots get closer to humanness with slight differences, users may experience discomfort, known as the “uncanny valley” phenomenon [22]. Inconsistencies in expressions, timing, or behavior can trigger these reactions [19]. Moreover, users often project their own traits onto robots, influenced by demographics and technical familiarity [2, 17]. These findings highlight the need to design robot personalities that are both context-sensitive and adaptable to diverse users, a consideration central to our study of QTrobot as a teaching assistant, where both social rapport and task effectiveness are essential.

Two participatory design (PD) workshops [23] were conducted in two days in the computing sciences department, each involving four master's students (N=8), recruited through snowball sampling. In the PD approach, participants acted as co-designers, giving ideas and feedback to improve QTrobot's personality trait perception and UX. Each workshop began with an introduction to the study and to QTrobot, an expressive, little humanoid social robot designed by LuxAI, Luxembourg. Participants completed three visual canvas tasks and interacted with QTrobot, who taught a “Social Media Ethics” lesson [24], which was programmed in QTrobot Studio incorporating facial expressions and upper-body gestures to convey emotions and verbal and non-verbal communication. The visual canvas tasks gathered participants’ expected personality traits (RQ1), the design elements and behaviors that conveyed personality traits (RQ2), and suggestions to improve the perception of personality traits. Then, participants completed an online questionnaire on their emotional responses during the interaction (RQ3) and a post-task interview to capture any missing insights on expectations, emotions, perceived personality traits, and improvement suggestions.

This study utilizes a combination of qualitative and quantitative data collection methods, including visual canvas tasks, post-task questionnaires, and post-task interviews. Three visual canvas tasks, implemented in Mural, were used to gather participants’ expected personality traits of QTrobot (RQ1), the robot's design elements and behaviors helped to identify personality traits during the interaction (RQ2), and design suggestions to improve the perception of personality traits of QTrobot. The post-task questionnaire assessed engagement, emotional reactions (RQ3), and perceived robot characteristics. It includes seven statements on user engagements (5-point Likert scale) adopted from “The User Engagement Scale” [25], fifteen rating statements from the “Godspeed” questionnaire [26] on the robot's anthropomorphism, animacy, likability, and perceived intelligence, five statements (5-point Likert scale) on emotional reactions of the participants, and two open-ended questions on participants’ emotional reactions and suggestions to improve. The post-task interview further explored participants’ experiences, expectations, and design suggestions. All sessions were audio-recorded and transcribed for analysis.

The gathered data were then analyzed using thematic analysis, affinity diagramming, and statistical methods. Thematic analysis [27] was used to identify themes and codes indicating key aspects from descriptive user feedback gathered through interviews and open-ended questionnaire responses. A Microsoft Excel workbook was used for thematic analysis, and derived themes and codes were then organized in the affinity diagram for further analysis based on research questions (RQs). The affinity diagram [28] served as the main qualitative data analysis method in this study, and it analyzed data gathered from visual canvas tasks, post-task questionnaires, and post-task interviews. Gathered data were entered into individual sticky notes on a Mural Canvas, each of which contains a single aspect. The sticky notes were placed under three main clusters: (1) personality traits expected from QTrobot as a teaching assistant (RQ1), (2) design elements and behaviors that helped to identify personality traits (RQ2), and (3) participants’ emotional reactions when interacting with QTrobot (RQ3). Basic statistical methods in Microsoft Excel were used to analyze quantitative data gathered from post-task questionnaires. Mean and mode values for each statement were calculated and utilized in interpreting the results of the questionnaire.

The study followed the General Data Protection Regulation (GDPR) to protect participants’ privacy. Informed consent was obtained, detailing the study procedure, data management, and rights to withdraw. To protect the participants’ privacy, identifier data, such as names, are replaced with unique codes (e.g., P#). All data were securely stored on a password-protected institutional OneDrive, accessible to researchers only. The consent forms and transcriptions were destroyed after completing the study. Participation in the study was voluntary, with the right to withdraw at any time without justification.

University students expect social, supportive, and professional traits while avoiding certain negative traits from QTrobot as a teaching assistant. Social traits such as sociability, friendliness, humor, happiness, politeness, interactivity, extroversion, and talkativeness were favored most, especially when conveyed with expressive facial gestures, speech, and movement. These traits helped the robot to maintain attention and encourage participation, especially among shy students. Supportive traits such as cooperation, guidance, kindness, and motivation were emphasized, with trust considered important, which was built through consistent, reliable, and accurate behavior. Students preferred QTrobot to convey warmth and not a machine-like tone, creating a welcoming atmosphere. Professional or cognitive traits such as adaptability, being organized, positivity, being careful, curiosity, and self-discipline were valued to encourage critical thinking and classroom balance. While most negative traits, such as aggression and depression, were rejected, some participants found occasional annoyance acceptable, viewing it humorously and positively. QTrobot's childlike appearance led to expectations of corresponding behaviors, and students preferred it to understand jokes without losing focus on lesson goals.

Design aspects and behaviors affecting students’ perceptions of QTrobot's personality traits were classified into three categories: nonverbal communication, language, and appearance. Nonverbal communication included facial expressions, body language, gestures, and voice. Smiling, waving, clapping, and expressive eye movements such as winking and raised eyebrows helped to convey friendliness and engagement, although some facial expressions and body language were confusing or unclear. The robot's voice and intonation conveyed a warm and cheerful atmosphere, but its childlike, high-pitched tone reduced perceptions of trustworthiness and respect. Language played a key role in conveying traits such as positivity, politeness, patience, and smartness with encouraging statements and detailed explanations. However, using too much smart or childlike language sometimes reduces trust. Similarly, QTrobot's childlike appearance reduced perceptions of reliability and seriousness among adult users. Questionnaire results (N=8) showed limited anthropomorphism but high ratings on liveness, interactivity, and responsiveness (mean>3, mode=4), indicating positive engagement. QTrobot was also scored highly on the likeability dimension (mean>3.5, mode=4) but perceived as less intelligent or competent (mean<3, mode=3). However, knowledge and responsibility were rated high (mean>3, mode=4), highlighting its potential as a social agent, but with areas to improve in perceived intelligence and anthropomorphic traits.

Emotional reactions and engagement were assessed by questionnaire (N=8), suggesting that QTrobot positively contributed to the learning experience. Participants felt supported, comfortable, and confident during interactions (mean>3, mode=4), with low anxiety about making mistakes. The robot successfully captured attention and facilitated immersion, with high rates of enjoyment, interest, and effectiveness. Negative experiences such as discouragement or frustration were given low scores (mean<2.5, mode=2), reinforcing QTrobot's contribution in providing positive, engaging, and low-stress educational interactions. Qualitative data from interviews and open-ended questions showed a mix of emotions. Positive responses included engagement, interest, and perceptions of “cuteness,” driven by interactive behaviors (e.g., clapping) and expressive features (e.g., eye blinks, raised eyebrows). However, negative emotions such as boredom, confusion, and frustration resulted from limited responsiveness, excessive speech, unnatural behaviors, and emotional disconnection. QTrobot's pre-programmed nature and robotic traits reduced its perceived lifelikeness. Participants recommended improving voice tone, expression variety, gesture rate, and real-time responsiveness to enhance emotional connection and natural interaction.