Pediatric Eye Exam Robot

Pediatric Eye Exam Robot

A 6-year-old grips her mother's hand. A mechanical arm extends toward her face. My job was to give that arm a soul before she ever walked into the room.

The Problem

The mechanical design was finalized. But the robot had no name, no face, and no way to tell a frightened child that everything was going to be okay. 53% of caregivers said the standard OCT exam is uncomfortable for children.

My Role

I owned the character design system, survey research and analysis, storyboard, and pre-exposure media. I translated raw caregiver data into a design direction the team could build from.

Role
UX Design Research Specialist
Team
UARTS Faculty Engineering/Arts Student Teams (FEAST) · Kellogg Eye Center
Timeline
Nov 2024 – Dec 2025
Tools
Figma · R · Procreate · ACM LaTeX

Key Design Decisions

Three choices that shaped the interaction system, each one traceable to survey data.

01

Character over function

The robot's mechanical capability was already defined; the missing layer was personality. Children meet Sam at home first, so by the time they walk into the exam room, he's someone they already know.

02

Data-driven design

I designed the survey as forced-choice pairwise comparisons and fit a Bradley-Terry model to the results. The model favored Sam in every matchup; the gaps weren't individually significant at n=30, so I triangulated with qualitative coding before recommending him.

03

Pre-exposure before first contact

Familiarity is built at home, before the child ever sees the arm. The pre-exposure video does that work.

The Robot

The hardware was done before I arrived. Everything the child would feel about it wasn't.

robot arm video demo

01

Research & Strategy

Four methods, one question: what actually makes children less afraid?

I started by shadowing real OCT procedures at Kellogg Eye Center: wait times, friction points, how children reacted to the equipment in the moment.

It's a super cool project. Has the potential to make the situation so much easier for children.

Caregiver, Kellogg Eye Center survey

Key insight

The model favored Sam in every matchup: 62% over Optic, 68% over Cheeze, 76% over A-Eye. At n=30 no gap cleared significance, but the direction never flipped.

The call

I recommended Sam on converging evidence, not one clean statistic, and documented the uncertainty in the paper.

Why forced-choice

Caregivers rate everything positively on a Likert scale. Forced choice surfaces what they'd actually pick.

Methods note: the IRB clock

Direct child surveys needed an IRB approval cycle longer than the project window, so I used a caregiver-report instrument instead, capturing children's reactions through parent observation. Their spontaneous reactions during sessions entered the data anyway.

Shadowing revealed two different kids in the same room, and two different design problems.

Age 6 Scared

"What is that thing? Is it going to hurt me?"

Needs comfort and familiarity before anything else.

Age 10 Bored

"How long is this going to take?"

Needs engagement and a sense of mission to stay cooperative.

Persona: Age 6, Scared
Persona: Age 10, Bored
personas

The Journey We Set Out to Change

How comfort shifts across the visit, with and without Sam.

Where the exam succeeds or fails Comfortable Anxious Home Waiting room First contact During exam Goodbye
Without Sam With Sam

Without Sam, comfort collapses at first contact. With Sam, it holds, which is what makes the scan possible.

Age 6, comfort first

Journey Map: Age 6

Age 10, mission first

Journey Map: Age 10
full journey maps from research

02

Character Design

Four concepts, one recommendation: Sam the Ranger, for cross-age appeal and clinical fit.

Each concept targeted a different emotional register. Survey data and medical staff review narrowed the field to Sam, a mission-driven ranger who reads well across ages.

One clear finding: children saw through A-Eye's staring-contest narrative and stopped cooperating, which is why the robot never lying became a first principle for the whole interaction system.

Character Design Board
character design board created with team
Optic Character + verdict

Optic

Strongest runner-up, especially with older curious kids. The mysterious persona needs pre-exposure videos, so it stays on the roadmap.

SAM the Ranger ★ prioritized

Sam the Ranger

The Bradley-Terry model predicts caregivers choose Sam head-to-head: 62% over Optic, 68% over Cheeze, 76% over A-Eye. One risk, the camera reading as a weapon, I reframed as binoculars in Sam's narrative.

A-Eye Character + verdict

A-Eye

Lowest scores, and the clearest lesson. Its narrative called the scan a staring contest, children caught the lie, and that failure became the system's first principle.

Cheeze Character + verdict

Cheeze

Strong visuals and storytelling, with a weak signal among anxious children. A photo interaction raised privacy flags, so future versions use pre-existing imagery.

four character finalists with Optic · Sam the Ranger · A-Eye · Cheeze

Every concept also had to survive the clinic itself: infection control, cleaning cycles, constant equipment contact. The fix was one robot with multiple swappable characters, outfits and masks that come off and go through the wash, so hygiene never had to compete with personality.

03

Storyboard & Pre-Exposure Media

I structured the storyboard to mirror the child's own emotional arc: Sam's introduction builds trust before the exam, and Sam's farewell closes the story with a good memory. I illustrated the pre-exposure video in Procreate for home and waiting-room viewing.

SAM the Ranger Storyboard
SAM the Ranger storyboard

04

Data Analysis & Deliverables

What the numbers said, what caregivers said, and what I did with both.

I built or analyzed everything below except the physical prototype, which the engineering team continued after my involvement ended.

Survey data analysis
survey data analysis

Survey Results

Method

30 caregivers completed a 40-question survey after viewing pre-exposure videos at Kellogg Eye Center.

Key insight

Over 60% reported curiosity and interest in every character; 80% rated Sam very or extremely effective at engaging children.

70%

of caregivers preferred character-guided interaction over the traditional pediatric OCT exam

4

robot character concepts designed and tested

2

personas from in-clinic observation

Head-to-head predictions

I fit a Bradley-Terry model to the forced-choice data to convert raw picks into matchup probabilities: 62% over Optic, 68% over Cheeze, 76% over A-Eye.

05

Interaction

The robot must never lie · Familiar before first contact · One robot, many characters

Every behavior below traces to one of these three principles.

The project itself ran about a year. After it wrapped, I kept refining this interaction system on my own.

Interaction Specification

Sam the Ranger service blueprint: journey phases (Pre-Exposure, First Contact, Calibration, The Game, Success & Goodbye) mapped across child actions, Sam's front-stage dialogue, and the system's back-stage mechanism for each phase.
Sam the Ranger service blueprint, exported from Figma

The blueprint above defines the system; the film below visualizes it.

A concept film visualizing Sam's behavior specification, from first contact to graceful failure recovery. AI-accelerated production; every shot maps to a spec row.

When it goes wrong

A spec that only covers the happy path isn't a spec. These are the cases caregivers and the clinical team actually raised: the moments a frightened child startles, cries, or moves during the scan, and how Sam responds.

Assumes voice output, a display, and basic motion/audio sensing. Detection methods are proposals for engineering validation.

Situation Detection (proposed) Sam's response Escalation
Child startled by Sam's approach Sudden recoil / gaze lost as the head nears Stop, ease back, lower the head (apology), then one reassuring nod Re-approach slowly; if still uneasy, pause and wait
Child cries Audio level Pause dialogue 5s, soften posture One gentle re-invite, then clinician handoff
Child moves during scan Motion exceeds threshold Calm reminder to stay still Pause scan, resume from last segment

Sam never

  • Sam never lies about the procedure. (From research: a deception-based concept was rejected on trust grounds.)
  • Sam never blocks the child's path to their parent.
  • Sam never raises volume or speeds up when a child disengages.

06

Next steps

My involvement ended December 2025.

Continued

The engineering team carried the physical prototype forward from there.

07

Takeaways

1
Emotional design is clinical design

In pediatric healthcare, a child's emotional state decides whether the exam can happen at all: a scared child moves, cries, or refuses, and the scan fails. Emotional design was the difference between a usable device and an unusable one.

2
Children are the sharpest honesty detectors

A-Eye failed because it lied, and that failure became the system's first principle. Children don't need the truth softened into fiction; they need a real role in a real story.

Published

A co-authored abstract in Investigative Ophthalmology & Visual Science, presented at the 2026 ARVO Annual Meeting.