lumaSpatial content lab
Interactive concept

Experiment 01 / Adaptive presence

Less interface.
More world.

An AR assistant that knows when to speak,
when to show, and when to step back.
Walk to the station. See the content adapt.

Glasses simulationFINAL CONCEPT
Policy simulator · No live sensors or AI
SPATIAL ASSIST · WALKINGDEMO

Station entrance

40 m · on your right

Mapping the space around you
Route labels placed · simulated
Urban scene: Unsplash ↗Conceptual placement

Voice transcript · reviewer view

“The station entrance is 40 metres ahead, on your right.”

British female preferred · loading available voices

01 /

Attention is the constraint.

Change the amount of content before changing the words.

02 /

Confidence changes placement.

Unreliable tracking removes anchors, not user control.

03 /

Rules before generation.

Approved recovery copy stays outside the generative layer.

Concept & content design Jay SpeakmanIndependent portfolio exploration · Not affiliated with GoogleSpatial / Multimodal / Adaptive
The content system

Design the conditions.
Then design the words.

Luma explores one task: finding a station entrance while an AR assistant adapts to attention and tracking constraints. The same intent survives as the delivery changes.

Separate policy from language.

A deterministic policy selects the channel, placement, and approved message. The prompt shown in the log is an illustrative contract for a future AI layer; this prototype makes no model calls. Critical recovery instructions use fixed, authored strings.

Fail visibly, recover deliberately.

Tracking loss removes all world-anchored content. Recovery first asks the user to stop safely, then find a well-lit surface. The final action explicitly simulates tracking recovery.

Make the tradeoffs reviewable.

The walkthrough connects navigation, tracking loss, recovery and arrival. Compare audio-only delivery with spatial cues, pause or repeat guidance, and simulate unavailable audio. Recovery instructions remain visible in either delivery mode. Driving is an optional policy stress test. The transcript and controls belong to the reviewer view. Reduced motion and keyboard controls are built in.

A browser concept, not a validated headset system. Scene mapping, distances, tracking and gaze are simulated. Real deployment would require hardware testing, attention and accessibility research, localization review, and safety validation. No microphone, camera, location, or gaze data is collected. Voice preview uses your browser's speech service. It prefers a British female voice when available. Voice names are used as a selection heuristic; the browser provides no gender or age property. Age and timbre cannot be guaranteed. A fixed recording would give a consistent voice across devices.