Built for Apple Vision Pro
SpatialRehab
Errorless-learning cognitive rehab for early-stage dementia, built for Apple Vision Pro.
- visionOS 2.0+
- SwiftUI
- RealityKit
- Swift Charts
Design overview · August 2026
The problem
Dementia takes the everyday first.
For someone in moderate cognitive decline, the hardest part of the day is the ordinary part. Long before anything dramatic, the small skills that hold a day together begin to fray. And it is not a small room.
Singapore · dementia prevalence
1 in 11
Singaporeans aged 60 and over live with dementia, and the share climbs steeply with age.
Projected · by 2030
152,000
People living with dementia here by 2030, up from about 74,000 in 2023.
Family caregivers
Say they are overwhelmed by what the role asks of them.
Routines
The steps of a familiar morning start arriving out of order, or not at all.
Wayfinding
A walk taken for thirty years suddenly stops feeling familiar.
Identity
Faces stay warm, but the names attached to them start to slip.
Sources: Well-being of the Singapore Elderly (WiSE) study, Institute of Mental Health, 2023; Ministry of Health projection, 2021; Dementia Singapore and Milieu Insight caregiver survey of 1,500 people, July 2023.
Why it matters
The goal is not a cure.
It is keeping the everyday doable.
Functional independence is the thing worth protecting: finding the way home, recognising a face, getting through a morning without someone else holding the steps together. This is a place to keep practising exactly those.
Low risk
A route can be walked, and mis-walked, from a chair in the living room. No traffic, no kerbs, no fall on an unfamiliar pavement.
Low worry
There is nothing here to fail. No score reaches the patient, so practice never becomes one more thing to be anxious about, for them or the family.
It transfers
Cues fade across repetitions on purpose, so the skill ends up with the person rather than with the headset.
Why this matters now
Singapore is projected to become a super-aged society in 2026, and by 2030 one in four citizens will be aged 65 or above. Anything that helps someone keep managing their own day, at home, matters more every year.
Ministry of Health, Singapore.
Our clinical foundation
Errorless learning.
The patient is never set up to fail.
Every screen in SpatialRehab is checked against one principle from dementia-rehabilitation research: support arrives before a mistake can happen, and it steps back as confidence returns.
Cue first
Prompts appear before a wrong answer is possible, never as a correction after one.
Reinforce, never punish
No red marks, no failure states, no score ever shown to the patient. Even a tally can read as pass or fail.
Fade gradually
Cues reduce over spaced repetitions, so ability transfers to real life instead of the headset.
The product
One calm window. One clear next step.
- A single window, never a wall of choices. One activity is offered at a time.
- Short sessions of 20 to 30 minutes, in line with Apple’s headset guidance.
- The window steps aside during immersive exercises, so the room itself becomes the interface.
Good afternoon.
TODAY’S CHECK-IN
Remember the Way
Study a route through the neighbourhood, then walk it from memory.
First run
Eight small games. One honest baseline.
Before anything else, a short run of eight games gently measures where the patient is today, so every later session has something real to compare against. Played in this order, warm-up first.
Reaction Time
Tap the shape the moment it appears.
Orientation
Day, month and time of day, asked live and never from a script.
Word Memory
Study a short list, then recognise it in a grid. No typing.
Digit Span
Watch four digits flash, tap them back in order.
Pattern Matching
Classic memory pairs. Mismatches simply turn back over.
Trail Making
Connect scattered numbers in order, at your own pace.
Arithmetic
Single-digit sums, tap the answer, one at a time.
Clock Drawing
Draw a clock face freehand. Kept for caregiver review, never auto-scored.
Results are stored for caregivers and the recommendation engine. The patient sees encouragement, never a number.
The first activity
Mahjong, by the real rules.
Singapore mahjong on a proper four-seat table against three computer players, with Pong, Chow and Mahjong claims live on every discard. The rules are not simplified. The deal is.
Fourteen tiles and the win. The seat on her left discarded the nine she was waiting on; the Chow claim was live only because that discard was genuinely legal.
A hand already nearly there
The deal is a genuine near-win and the wall is kind, so the game stays reachable.
Opponents that feed
The computer players put out tiles she can use, and a claim button appears only when the claim is legal.
No score at the table
Turns, claims and misdrops are counted quietly for the caregiver view. What she sees is the win.
An everyday sequence
Making kopi, one step at a time.
An activity of daily living, hand tracked on a real tabletop: procedural memory and task sequencing, practised in the order she would actually do it.
Before anything is poured, she chooses how much help she wants. Nothing is highlighted or demonstrated until she has picked, and “Make another” asks again, so the level can come down as she grows more sure.
Show me how
“I’ll talk you through every step and show you how it’s done.” Every cue arrives at once, the demonstration within two seconds.
A little hint
“I’ll point out what’s next, and show you if you’d like a moment.” The cues join in from five seconds, the demonstration at twenty.
Let me try myself
“I’ll stay quiet, and step in only if you’d like a hand.” Twelve seconds of silence, and only the step is named, never the recipe.
The levels fade the cues, never the task: every level still ends up offering the full demonstration, and the clock restarts the moment she picks the right thing up. On the table, amber is the thing to pick up and cyan is where it goes.
Identity
“Who am I?”
Always one tap away.
A floating card she can open from anywhere in the app: her photo, her name in English and Chinese, when she was born, where she lives, who to call. The things a person reaches for when the ground moves under them.
It is built to read as her own keepsake rather than a test. Layered glass, entrances that arrive slowly and springily, a sheen that sweeps once and stops. Nothing on this card is ever asked back.
Lim Chio Bu
林招母
This is me
The people who matter
The card flips, and the family is there.
Six people, laid out as a tree whose every line is drawn between the real, measured positions of the tiles, so no one floats unconnected and each name has a line physically touching it. She is marked at the centre of her own family.
“Brian usually buys me Min Jiang Kueh on Saturdays.”
Every person carries a note in her own words. Writing those notes herself is the next step: her context, her reminders, whatever makes this face familiar again.
Recognition
A pinch on a face,
and they speak to her.
Every face in the tree is already moving: a muted clip looping behind the circle, so the family reads as people rather than a grid of photographs. Pinch one and the card turns to a greeting recorded by that person, played once, and then she is handed back to the tree.
- Six loops at once. One retained looping player per portrait, built once and reused, because a player rebuilt on every re-render would throttle the device within a minute.
- Quiet until asked. The portraits are muted and carry no controls. Sound arrives only when she chooses a face.
- Nothing is asked back. No name to recall and no button to get right. She watches, it ends, and the card returns on its own.
Lim Ah Pek
Her husband · 您的丈夫
“Ah Bu, come drink kopi with me. · 招母，来喝杯咖啡。”
The flagship exercise
Remember the Way.
A real OpenStreetMap extract of the user's neighbourhood, procedurally meshed at runtime, with shophouses, road ribbons, street lamps and trees, then mounted as a miniature on a holographic tabletop.
Real building footprints and street geometry, drawn from the same OpenStreetMap extract the app meshes at runtime. Map data © OpenStreetMap contributors.
The route draws itself
The path animates along the tabletop and repeats, so it has a chance to encode.
Trace it from memory
The map locks and the person draws the route back. Scored by distance from the real polyline, never shown as pass or fail.
Then step inside
The map scales up around her.
The same estate and the same route, at life size. She stays seated. A pinch held between finger and thumb carries her forward along the path, and letting go stops her. There is no stick to steer and no direction to get wrong.
One slow, constant speed
Velocity eases in and out over about 150 milliseconds and never varies with how hard she pinches. Small heading changes snap invisibly; sharp corners take a blink cut rather than a swung camera.
A voice the whole way
Spoken directions travel with her, and the arrival is said out loud: “You’ve reached home. Well done.”
It ends at her front door
The walk finishes where the route finishes and nothing restarts on its own. Stopping is only ever letting go.
For caregivers
Progress you can actually see.
- Every session is kept, not overwritten, giving a real history to compare against the baseline.
- One trend line per game, plus a gallery of every clock drawing over time.
- Raw data sits one tap deeper. The dashboard reads plainly, and the detail is there when a clinician wants it.
Word memory · weekly score
Sample data, for illustration.
Adaptive by design
It notices what’s hardest,
and meets the patient there.
Each game maps to a cognitive domain. The recommendation engine ranks domains by measured weakness and suggests what to practise next. It is a plain, explainable calculation, because a caregiver deserves to see exactly why.
Not a black box: no trained model, no hidden weights. The weakest domain surfaces first, and the reasoning is visible at a glance. Values shown are illustrative.
Evidence-based
Calibrated to the research,
not to the demo.
Session length, cadence and programme duration are taken from published dementia-intervention findings and Apple’s own guidance on headset comfort. Every number here has a source behind it.
Cognitive stimulation, the core of every session
20–30 min
Per session, within Apple’s guidance for continuous headset use.
3× / week
Or more. The intervention cadence with the strongest measured outcomes.
5–8 weeks
Programme length where the evidence is strongest. Longer isn’t necessarily better.
Seated physical activity Planned
Chair-based movement, marching and arm raises, is a safe way to raise activity for someone at risk of falls, and physical activity carries some of the strongest evidence for slowing decline. The literature points to 3 to 4 times a week, 30 to 45 minutes. It sits alongside the cognitive work rather than replacing it.
A hackathon prototype, not a medical device. Not clinically validated, and making no diagnostic claims. Those eight games borrow structure from established concepts such as clock drawing, digit span and orientation; their specific content is still pending clinical review.
The team
Five people, each owning a slice.
Aditya
3D Assets & Environment
Interactive 3D models and scenes.
Brian
UX/UI & Spatial Design
Onboarding, guidance, interaction flow.
JingTong
Apple Vision Pro Development
RealityKit scene setup and hand tracking.
Nicole
Clinical Research & Product Management
Evidence-based exercise design and validation, plus product direction.
Basil
AI, Analytics & Integration
Baseline metrics, recommendation engine, caregiver dashboard.
A rehearsal space for a mind, and for the route home.
Practice makes the route feel like home again.
Thank you.
Photography courtesy of the Ministry of Health and Age Well SG, Singapore