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

This exploratory study with 12 elementary students demonstrates that socially assistive robots can effectively facilitate peer mediation role-play for conflict resolution, yielding positive student feedback and revealing significant correlations between trait and learning measures in the robot condition despite the lack of statistical differences compared to a tablet-only control.

The Gist

Imagine a playground where two kids are arguing over a toy. Usually, a teacher steps in to fix it. But what if the kids could learn to fix it themselves? That's the idea behind Peer Mediation, a program used in many schools where students help other students resolve their fights.

However, not every school has a program like this. That's where this research comes in. The team from the University of Southern California asked a big question: Can a robot help a child learn how to be a "peacekeeper" for their friends?

Here is a breakdown of their study, explained simply with some creative analogies.

The Experiment: A "Playground" for Robots and Kids

The researchers set up a little "conflict simulation." They wanted to see if kids could practice being mediators (the peacemakers) by talking through a pretend argument.

They split 12 elementary school kids into two groups:

1. The Robot Group: The kids sat in front of two small, cute, flower-shaped robots (named "Blossoms"). These robots acted out the roles of the two kids fighting.
2. The Tablet Group: The kids sat in front of a tablet screen. The "fighting kids" were just voices and pictures on the screen.

The Mission: The student had to read a script, act as the referee, and help the two "fighters" (whether they were robots or screen characters) make up and solve their problem.

The "Magic" of the Robots

Why use robots? Think of robots like a judgment-free practice partner.

• If you practice talking to a real person, you might feel shy or worried about saying the wrong thing.
• If you practice with a robot, it's like talking to a friendly video game character. It's safe, fun, and doesn't get mad if you stumble over your words. The researchers hoped this would make the kids feel more confident.

What They Found (The Good, The Bad, and The Surprising)

1. The Kids Loved It (The "Fun" Factor)
Almost every child said the activity was fun and helpful. Many told the researchers, "I feel better about myself now," or "I learned how to help my friends." It was like a mini-gym class for their social skills, and they enjoyed the workout.

2. The Reading Hurdle (The "Speed Bump")
Here was a big snag: The activity required the kids to read a script out loud.

• The Analogy: Imagine trying to learn how to drive a car, but the steering wheel is made of heavy lead. The kids' reading skills varied wildly. Some read fast and smoothly; others struggled to read the words.
• The Result: This made it hard to tell if the robots were the reason for the success, or if the kids who read well just did better because they could read the script easily. The reading difficulty was a "confounding variable" (a fancy way of saying a messy factor that muddied the water).

3. Robots vs. Tablets: The "Tie"
When the researchers looked at the test scores and how the kids felt about themselves before and after the activity, there was no huge difference between the Robot group and the Tablet group.

• It wasn't that the robots failed; it's that the study was too small and short to see a clear winner. It's like flipping a coin twice and getting heads both times—you can't say the coin is "lucky" yet; you need more flips.

4. The Secret Connection: Personality Matters
This is the most interesting part. While the average scores were the same, the researchers found some hidden patterns in the Robot group specifically:

• The "Chatty" Kids: Kids who were naturally more outgoing (extroverted) took longer to read the script when talking to the robots. It's like they were so engaged in the "conversation" with the robot that they paused to think, "Wait, what does this robot want me to say next?"
• The "Careful" Kids: Kids who were naturally more careful and detail-oriented (conscientious) took their time and did well.
• The "Nice" Kids: Kids who were naturally very agreeable (friendly/cooperative) guessed the right answers faster in the robot group.

The Takeaway: The robots seemed to interact differently with different personality types. The tablet was just a screen, but the robot felt like a character, which made the kids' natural personalities shine through in how they played the game.

What's Next?

The researchers realized that for this to work really well, they need to:

• Fix the Reading: Make the activity less about reading words and more about listening and speaking, so kids of all reading levels can play.
• Play Longer: Do this over several weeks, not just one afternoon. Changing how a child feels about themselves is like growing a plant; you can't do it in 30 minutes.
• Watch the Kids: They plan to use cameras to see how kids move their bodies and look at the robots, to understand why the robots made some kids act differently.

The Bottom Line

This study was a "test drive." It showed that using robots to teach kids how to solve conflicts is a promising idea. The kids had fun, felt good about themselves, and the robots didn't scare them. While the robots didn't magically make everyone smarter than the tablets in this short test, they did reveal that robots might be special because they react to a child's personality in a unique way.

In the future, these robots could be like digital "peers" that help kids practice being kind, patient, and good listeners before they have to do it in the real world.

Technical Summary

Here is a detailed technical summary of the paper "Exploring Socially Assistive Peer Mediation Robots for Teaching Conflict Resolution to Elementary School Students."

1. Problem Statement

Conflict resolution is a critical life skill, and peer mediation is a widely used educational model in K-12 schools where students help resolve conflicts among peers. However, many schools lack the resources or programs to implement peer mediation training.

• The Gap: There is a need for accessible, scalable tools to practice these skills, particularly for schools without existing programs.
• The Hypothesis: Socially Assistive Robots (SARs) can augment social-emotional education. While previous research has explored robots acting as mediators or tutors, this study investigates a novel approach: using SARs as mock disputants (the parties in conflict) to allow human students to practice the role of the peer mediator.
• Research Questions (RQs):
  1. How do elementary students respond to peer mediation with SARs compared to a tablet-only interface?
  2. Are there differences in self-perception or learning outcomes between the two conditions?
  3. Are participant behavioral traits (personality, conflict style) associated with self-perception changes or learning efficacy?

2. Methodology

Study Design

• Type: Exploratory, single-session, between-subjects user study.
• Participants: $N=12$ elementary school students (Grades 4–5, ages 9–11).
  • Demographics: 5 Male, 7 Female; 11 Hispanic/Latino, 1 White.
  • Assignment: Randomly assigned to either the Robot Condition ($n=6$) or the Tablet Condition ($n=6$).
• Setting: A local elementary school (St. Odilia School).

Experimental Conditions

1. Robot Condition: Participants interacted with two Blossom robots (3D-printed, open-source, zoomorphic design) acting as the disputants ("Nova" and "Dali"). The robots were controlled via Raspberry Pi and displayed movements synchronized with speech.
2. Tablet Condition: Participants interacted with a tablet interface where the disputants were represented only by pre-recorded voice audio and static images.

• Common Elements: Both groups used a tablet UI to view scripts, select responses, and receive feedback.

The Task: Peer Mediation Role-Play

• Script: Based on a script from the Conflict Resolution Center of St. Louis, adapted for age-appropriateness.
• Mechanism: The student acted as the mediator. The script included:
  • Dialogue: The student read lines aloud; the "disputants" (robots or tablet) responded.
  • Decision Points: At specific turns, the student had to fill in blanks by choosing from three multiple-choice options to guide the conflict toward resolution.
  • Feedback: If an incorrect choice was made, the system provided indirect hints (up to 3 times).
• Data Collection:
  • Pre/Post Surveys: Self-Perception Profile for Children (SPP-C) for behavioral conduct; How I Respond to Conflict Worksheet (RCW).
  • Knowledge Check: A 5-question quiz on peer mediation concepts (pre- and post-session).
  • Personality Assessment: Pictorial Personality Traits Questionnaire for Children (PPTQ-C) based on the Big-Five framework.
  • Behavioral Metrics: Duration of turns, number of attempts to answer correctly, video/audio recordings of facial expressions and body language.

3. Key Contributions

1. Novel Interaction Paradigm: This is the first study to utilize SARs specifically as mock disputants in a role-play setting to teach peer mediation, shifting the robot's role from tutor/mediator to the "problem" the student must solve.
2. Exploratory Correlation Analysis: The study moves beyond simple pre/post comparisons to investigate how personality traits (e.g., extraversion, conscientiousness) and conflict response styles correlate with learning metrics (time taken, number of attempts) specifically in the presence of embodied robots.
3. Identification of Design Constraints: The study highlights reading proficiency as a significant confounding variable in robot-mediated educational activities for children, noting that varying reading levels impacted the ability to follow the script and dictate turns.

4. Results

Qualitative Findings

• User Sentiment: The majority of students (11/12) reported the activity helped them feel better about themselves.
• Perceived Utility: Students expressed that the activity taught them how to help friends during conflicts.
• Engagement: Observations noted varying levels of expressiveness; some students were highly engaged and expressive, while others were reserved.

Quantitative Findings

• Self-Perception (SPP-C):
  • No statistically significant difference was found between the Robot and Tablet conditions regarding changes in self-perception ($p = 0.053$, close to significance).
  • The Robot condition showed a slight average decrease in scores, while the Tablet condition showed a slight increase.
• Learning Outcomes (Quiz Performance):
  • No significant difference in quiz score improvements between conditions ($p = 1.0$). Both groups showed a negligible average increase of 0.2 points.
• Trait vs. Learning Correlations (Robot Condition Only):
  • Significant correlations were found between personality traits and time taken and number of attempts in the Robot condition, but not in the Tablet condition.
  • Extraversion: Strong positive correlation with time taken on Turn 3 ($r=0.84$) and attempts on Turn 18 ($r=0.88$). Suggests extroverts took longer or struggled more with specific turns.
  • Neuroticism: Strong negative correlation with time taken on Turn 3 ($r=-0.86$).
  • Conscientiousness: Strong positive correlation with time taken on Turn 3 ($r=0.84$).
  • Agreeableness: Strong negative correlation with attempts on Turn 5 ($r=-0.90$), suggesting agreeable participants solved the recall question faster.
  • Conflict Response: Avoidant response preference correlated negatively with time on Turn 3 ($r=-0.85$).

5. Significance and Future Work

• Promise of SARs: While statistical differences in learning outcomes were not found (likely due to small sample size and short duration), the strong correlations between traits and performance in the Robot condition suggest that SARs interact with student personality in ways that tablet interfaces do not. This indicates that SARs may offer a unique, personalized learning dynamic.
• Design Implications:
  • Reading Barriers: Future iterations must address reading level disparities, potentially through voice-to-text or audio-only scripts, to prevent reading ability from confounding the mediation skills being tested.
  • Longitudinal Studies: Self-perception changes likely require longer intervention periods than a single 30-minute session.
  • Affective Computing: Future work should utilize the collected video/audio data to quantify expressiveness and engagement, linking these metrics to the observed trait correlations.
• Conclusion: The study validates the feasibility of using SARs as mock disputants for peer mediation training and provides a foundational framework for analyzing how student personality traits influence learning efficacy in embodied vs. non-embodied educational contexts.