OVERVIEW
What
User Experience Research Design to investigate American kids and teachers experiences with Haru in classroom and if this interaction has any impact on improving communication skills on school age kids.
Users
USA Elementary and Middle School
Kids and Educators.
Timeframe
10 weeks
Tools
Mural, Nvivo, Microsoft 365,
Indesign & llustrator
Team
Erin Clepper, Dr. Liz Sanders (OSU), Maryam Alihoseini, Priyanka Chowdhury, Heloisa R. Rincon, Howie Wang
My role
■ Led secondary research on best practices to design child-centred surveys/ assessments;
■ Led secondary research on best practices to validate multiple coder analysis
■ Designed Co-design workshops and focus groups in collaboration with the team
■ Designed assessment workbooks and tools kits;
■ Designed Induction Coding Map;
■ Designed thematic analysis workflow for multiple coders;
■ Mentored team on the use of Nvivo.
OUR VISION
Explore how children learn through educational technologies using human-centred design research to inform responsible human–machine experiences.
Meet Haru
Haru is a social robot, designed to foster emotional connections with humans. It has been designed, developed and intensely researched over the past 7 years by Honda Research Institute (HRI) Japan, through a multidisciplinary collaboration engineers, designers, academics, researchers, artists, and animators all over the world.
In 2025, a new phase of Haru Project started and HRI USA was invited to join contributing on AI-Human interaction user experience research with teachers and kids from elementary and middle school. For this new phase the goal was to investigate how Haru could foster the improvement of communication skills in schools
Challenge
How might we design human-robot interactions between local groups in unstructured pedagogical settings to address target children’s communication skills
?
The communication skills of school age children have been impacted by the recent global shutdown due to Covid.
1
Limited opportunities for practicing.
2
Need for constant guidance and feedback
3
Difficulty in expressing oneself
4
Shyness and social inhibition
5
Learned helplessness
Research Plan
I joined the project during its second phase for ten weeks, supporting focus groups sessions design, shaping the analysis plan, and conducting secondary research to inform decisions.
Below, I present four examples that I led or owned to demonstrate how my role contributed to the project.
Research Tools
For this study, we used a mixed-methods to collect qualitative and quantitative data during focus groups with children, the robot, and the teacher in a controlled environment. We designed self-report assessments, and planned video and audio recordings of interactions with Haru through the activities we created for the focus groups, as well as field notes.
This session highlights the process of designing the child-centered self-report assessments (1.0) and the ideation of focus group activities (2.0).
1.0 Child-Centred Self-Report Tool Design
Need to design an assessment tool (workbook) to complement observational tools during workshops.
Goal: Capture children’s authentic experiences in a structured, playful, and engaging way
Context
Survey challenge: Designing reliable, age-appropriate surveys is complex.
Child response: Younger participants may misinterpret, forget, or give answers to please adults.
Likert issues: Traditional likeart scales are often misused or biased.
Challenge
Direct child input: Children provide feedback adults cannot.
Survey design: Short, clear, age-appropriate, pretested.
Smiley scale: 5-level happiness scale reduces response bias.
Embodied surveys: Support playful, reflective feedback.
Insights
The Workbook
What it is: A child-friendly tool for ranking activities from “liked least” to “liked most” (adapted from Sylla et al., 2017, 2019).
How it works: Children place post-its with pictures representing the activities played during focus group on ladder steps, reflecting on preferences while researchers capture thinking and authentic feedback.
Why it matters: Makes evaluation playful, interactive, and reflective, while generating genuine insights into children’s preferences and thought processes.
Based on previous experiences working with children and on secondary research I designed a child-centered assessment workbook to replace traditional surveys and gather meaningful insights.
Here you can see three example of activities presented on the workbooks.
Relevance
I redesigned traditional survey instruments into child-centered multimodal workbooks to increase engagement, reduce interpretive bias, and strengthen the validity of qualitative and quantitative responses.
2.0 Focus Groups Activities Ideation
Grounding
■ We used secondary research and prior co-design workshops with local teachers to inform activity ideation.
■ Designed age-specific activities with separate matrices for elementary and middle school.
■ We made sure activities addressed communication challenges and considered interactions between the robot, teacher, and students.
I partnered with a teammate to ideate activities to the focus groups
Systematization
■ We designed a structured Mural matrix with metrics to classify and assess activity features, robot roles, and ecological feasibility.
■ We organized the large set of ideas into a clear, research-informed structure.
Impact
■The matrix enabled transparent prioritization and comparison of activities.
■ It supported team voting and teacher feedback
■ It served as a solid foundation to turn raw ideas into a defensible research plan.
Analysis Framework & Reliability
.png)
Data & Rationale
6 schools (3 elementary, 3 middle), 3 sessions each (18 total). 1 continuous group per school: 3–4 students + 1 teacher. Data: quantitative and qualitative, from workbooks, transcripts, and ~24 hrs of 360° videos from focus group sessions.
Code Map
We designed a initial deductive coding list that could capture emotional affect, participant roles, engagement, turn-taking, and conflict markers. I built a code map to show connections, guide code rules, and support real-time refinement of the codes.
Automation
The team explored automated transcription and sentiment extraction across text and audio. Outputs will be treated as probabilistic signals — not conclusions — to support, not replace, human interpretation.
NVIVO
All materials were organized in NVivo to enable thematic synthesis and support consistent analysis across researchers. I guided the team in using NVivo throughout the process.
Relevance
By visualizing the coding system, we could quickly spot gaps and refine codes in real time targeting an comprehensive codebook for the future
Designing Trust in Qualitative Analysis
During a test, intercoder agreement reached 70%, revealing ambiguity in the deductive coding system.