Exploring Socially Assistive Peer Mediation Robots for Teaching Conflict Resolution to Elementary School Students

AI summary

Overview

Research area: Human-Computer Interaction / Human-Robot Interaction, specifically socially assistive robotics (SARs) for children's social-emotional education.

Technical level: Intermediate. The domain concepts (peer mediation, role-play, self-perception scales) are accessible to a general reader, but the paper's analysis relies on t-tests, Mann-Whitney U tests, Pearson and Spearman correlations, and child-specific psychometric instruments (SPP-C, PPTQ-C, RCW).

Scope: A single-session, between-subjects exploratory study with 12 elementary school students comparing a peer-mediation role-play activity conducted with two Blossom SARs as mock disputants versus the same activity presented only through a tablet interface.

What This Paper Is About

Peer mediation is a conflict-resolution model used in many K-12 schools in the United States, in which students help other students resolve conflicts. Schools without such programs may lack an accessible way to let children practice these skills. This paper asks whether socially assistive robots can serve as mock disputants in a scripted peer-mediation role-play, and whether children respond differently to that activity when the disputants are embodied robots versus a tablet interface.

Key Contributions

1. A novel SAR role design. Instead of robots acting as mediators or tutors, the two SARs (Blossom robots) play the role of disputants in a conflict scenario, placing the child in the position of guiding them rather than being guided. The authors state this is, to their knowledge, the first use of SARs for role-play to teach students peer mediation.

2. A complete peer-mediation role-play system and script. The system combines two Blossom robots, a tablet UI with mouse and keyboard, text-to-speech child voices (Narakeet), on-screen turn-by-turn script display, embedded multiple-choice mediation questions with up to three attempts and indirect hints, and a score counter. The script was adapted from a pre-existing script by the Conflict Resolution Center of St. Louis to be age-appropriate.

3. Preliminary quantitative and qualitative findings from a 12-participant between-subjects study (6 robot, 6 tablet), covering self-perception change (SPP-C), quiz performance, personality (PPTQ-C), and conflict response preference (RCW).

4. Correlational evidence linking traits to learning-related measures in the robot condition only, using turn-level duration and number of attempts as learning-efficacy proxies. These correlations appear in the robot condition and not in the tablet condition.

Main Findings

• No significant embodiment effect on self-perception. The change in SPP-C scores (post minus pre) was M = −0.5, SD = 0.5 for the robot condition and M = 0.11, SD = 0.5 for the tablet condition, t(10) = −2.2, p = 0.053. The authors describe this p-value as close to the 0.05 significance level, indicating a slight possibility of a difference. In the robot condition there was a slight average decrease in SPP-C score; in the tablet condition a small average increase.

• No significant embodiment effect on quiz performance. The change in number of correct quiz answers was M = 0.2, SD = 0.7 for both the robot and tablet conditions, t(10) = 0.0, p = 1.0. In both conditions, three participants showed no improvement, two improved by one question, and one answered one additional question incorrectly.

• Positive qualitative feedback. When asked whether the activity made them feel better about themselves, eleven participants responded that it did, with three responding that it helped a lot and one responding that it helped a little bit. All responses to the open-ended question about the peer mediation activity were positive, mentioning it being fun and helpful. One participant wrote that they learned how to help friends because they see classmates argue and do not know how to handle it; another wrote about making new friends or helping them be friends again.

• Traits correlated with turn duration in the robot condition only. For turn 3 of the script, significant Pearson correlations were found with extraversion, r(6) = 0.84, p = 0.04; neuroticism, r(6) = −0.86, p = 0.04; and conscientiousness, r(6) = 0.84, p = 0.04. No significant correlations were found for the tablet condition.

• Traits correlated with number of attempts in the robot condition only. For turn 5, agreeableness showed a significant strong negative Spearman correlation, r(6) = −0.90, p = 0.01. For turn 18, extraversion showed a significant strong positive Spearman correlation, r(6) = 0.88, p = 0.02.

• Conflict response preference correlated with turn duration and attempts in the robot condition only. For turn 3, avoidant response preference showed a significant strong negative Spearman correlation, r(6) = −0.85, p = 0.03. For turn 5, cooperative response preference showed a significant strong negative Pearson correlation, r(6) = −0.87, p = 0.03.

• No trait correlations with self-perception change. No significant correlations were found between personality, conflict response preference, or prior SPP-C scores and SPP-C delta in either condition. The only near-significant result was conscientiousness in the tablet condition, r(6) = 0.79, p = 0.06.

• Reading level was a confounding factor. Participants had varying reading levels that affected their ability to read and dictate the turns in the role-play script. One third-grade participant had difficulty reading the script, required significant support from the researcher, and was omitted from the analysis. The authors note participants often read aloud slowly.

• Condition distributions were comparable at baseline. Personality, prior SPP-C, conflict response preference, and activity time showed no statistically significant differences between conditions. Measures reported include prior SPP-C (robot M = 3.1, SD = 0.4; tablet M = 2.4, SD = 0.6; t(10) = −2.1, p = 0.06) and activity duration in seconds (robot M = 604.8, SD = 162.6; tablet M = 496.7, SD = 135.0; t(10) = −1.14, p = 0.28), roughly 8–10 minutes.

• Expressiveness varied across participants. Some participants were outwardly highly engaged, curious, and expressive, while others were more reserved, uninterested, or bored. The authors report they are working on quantifying this using OpenFace and MediaPipe on facial and postural features.

Methodology in Plain English

The activity. Each child played the role of a peer mediator in a scripted conflict between two hypothetical students named Nova and Dali. The child spoke aloud in English, and the disputants responded either by speaking to the child or to each other. Every turn of the conversation was also displayed as text on the tablet UI, including the child's own lines. At certain points, part of the child's script was blank and the child chose among three answer options. A wrong answer triggered an indirect hint ("Let's think again, what would be the best way to fill in this blank?") for up to three attempts. A counter tracked correct and incorrect answers, and a summary was shown at the end.

The two conditions. In the robot condition, the disputants were two updated Blossom robots (3D-printed, open-source platform originally developed by Suguitan and Hoffman, further refined by Shi et al.), controlled by Raspberry Pi computers, with crocheted exteriors and button features, one blue and one gray to distinguish the two characters. In the tablet condition, the disputants were represented only through voice audio and on-screen images. The tablet was a Gechic portable 7.5 in. x 11 in. monitor with a keyboard and mouse.

Robot behavior. Blossom robots were programmed to display movements consistent with each robot's speech, such as strong head shakes for anger or slouching for sadness. Movements were pilot tested with 3-5th grade students in a university summer coding program. Idle movements such as simulated breathing or swaying were found to be distracting and were excluded, so the robots moved only while speaking.

Data collection. Two Logitech webcams captured data: one on a tabletop tripod across from the participant (face and speech audio) and one on a tripod to the participant's left (full-body profile).

Participants and procedure. The school was recruited via an informational flier. Eligibility was enrollment in 3-5th grade (ages 8-11), English fluency, and in-person participation. Eighteen students were recruited; two did not assent. Of the 16 who assented, one had a technical failure and was removed. Because the script was simplified after the first day, the two participants from that day were omitted. One third-grade student who required significant researcher support was also omitted. The final analysis is based on 12 participants: 5 Male, 7 Female; 11 Hispanic/Latino, 1 White; ages 9-11 (M = 9.9, SD = 0.6); grades 4th-5th; six per condition, randomly assigned. Three participants in the robot condition reported having seen or used a robot before, and one participant in the tablet condition reported familiarity with peer mediation. Total study duration took about 30 minutes (M = 34.5, SD = 4.5). The study was approved by the university IRB (#UP-24-00495).

Measures. Pre-session: the behavioral conduct subscale of the Self-Perception Profile for Children (SPP-C), the How I Respond to Conflict Worksheet (RCW), a short conflict resolution lesson, and a five-question quiz on peer mediation. Post-session: repeated SPP-C and quiz, the Pictorial Personality Traits Questionnaire for Children (PPTQ-C, 15 items based on the Big-Five framework for ages 7-13), plus demographics, prior experience, and open-ended feedback.

Analysis. Quantitative tests were run in Pingouin version 0.5.5. Independent samples t-tests compared pre-/post-session deltas in SPP-C and quiz scores across conditions, with Shapiro-Wilk tests for normality and Levene tests for variance equality. Correlations were computed between three trait-related measures (personality, conflict response preference, prior self-perception) and two learning-related measures: duration of turns (selected as the highest-variance turns in each condition) and number of attempts. Pearson was used for normally distributed variables and Spearman's rank for those that failed Shapiro-Wilk.

Why This Matters

Impact on research. The paper introduces a design pattern—robots as mock disputants rather than mediators or tutors—that inverts the usual SAR-as-guide arrangement. It also provides an early empirical signal that personality and conflict-response traits correlate with turn-level engagement measures in the robot condition but not the tablet condition, which the authors argue suggests personality may interact more strongly with embodied peers. The study also surfaces reading level as a design confound that future child-robot interaction studies should plan for.

Real-world applications:

• Schools without peer mediation programs could use SAR-based role-play as an accessible supplement or practice tool for conflict resolution skills.
• Peer mediator training, which commonly relies on role-play, could use robot disputants to provide consistent, judgment-free practice partners that behave identically across sessions.
• Social-emotional learning curricula in K-12 could incorporate short scripted robot-mediated scenarios as a structured activity.
• Child-facing educational technology design can use the finding that reading load and turn pacing confound outcomes to build lower-text or audio-supported interfaces.

Industry relevance. The work is relevant to developers of educational robots and children's conversational systems, particularly around embodiment choices (two-robot setups versus screen-only), pacing of turn-taking for young users with varying literacy, and the use of pre-scripted content rather than generative content. The authors deliberately used a pre-planned script rather than an LLM-generated scenario, explicitly to ensure student safety, which stands in contrast to LLM-based conflict-resolution training approaches such as Shaikh et al. (2024).

Future Directions

1. Reduce the reading barrier. The authors plan to limit the reading load so conflict resolution can be taught in a more personalized way, and to conduct co-design sessions and pilot studies with the intended participant population to diminish unexpected confounds.

2. Analyze affect and eye gaze. Using the collected video and audio data (with OpenFace and MediaPipe), the authors intend to quantify expressivity and engagement and test whether trait-related measures relate to participant behavior during the activity.

3. Run longer and larger studies. Because changes in self-perception are better suited to long-term studies, the authors state future peer mediation activities with SARs should ideally be implemented and evaluated over weeks of practice, and that personality's relationship with self-perception should be investigated in a larger-scale study.

4. Open questions the paper raises. Why do agreeableness, extraversion, and avoidant or cooperative response preferences correlate with comprehension and pacing in the robot condition but not the tablet condition? The authors suggest a possible confounding variable related to both comprehension and conflict response preference as interesting future work.

Target Audience

This paper is most useful to HRI and socially assistive robotics researchers studying child-robot interaction, education technology designers building conflict-resolution or social-emotional learning tools, and school practitioners or administrators considering peer mediation programs. It is also relevant to researchers designing child studies who need to anticipate reading-level confounds, small-sample exploratory designs, and the practical scheduling constraints of conducting research in elementary schools. Because the findings are preliminary and the sample is 12 participants across two conditions, readers seeking definitive evidence of robot superiority over tablet presentation will not find it here; the paper reports no significant differences on its primary pre-/post-session measures and frames its contribution as informing future study design.

Authors’ abstract

In peer mediation--an approach to conflict resolution used in many K-12 schools in the United States--students help other students to resolve conflicts. For schools without peer mediation programs, socially assistive robots (SARs) may be able to provide an accessible option to practice peer mediation. We investigate how elementary school students react to a peer mediator role-play activity through an exploratory study with SARs. We conducted a small single-session between-subjects study with 12 participants. The study had two conditions, one with two robots acting as disputants, and the other without the robots and just the tablet. We found that a majority of students had positive feedback on the activity, with many students saying the peer mediation practice helped them feel better about themselves. Some said that the activity taught them how to help friends during conflict, indicating that the use of SARs for peer mediation practice is promising. We observed that participants had varying reading levels that impacted their ability to read and dictate the turns in the role-play script, an important consideration for future study design. Additionally, we found that some participants were more expressive while reading the script and throughout the activity. Although we did not find statistical differences in pre-/post-session self-perception and quiz performance between the robot and tablet conditions, we found strong correlations (p<0.05) between certain trait-related measures and learning-related measures in the robot condition, which can inform future study design for SARs for this and related contexts.