Spatial Touchpad Interaction Model
Overview
Modern touchpads capture rich spatial data but reduce it to cursor movement. This project reclaims that lost layer — treating the touchpad as a 2D interaction surface with meaning tied to position, direction, and time.
We designed and built a working system where users navigate a document entirely through spatial gestures, without ever using a cursor.
Problem
Reading on laptops is fragmented:
Constant switching between keyboard, trackpad, and cursor Gestures are arbitrary and hard to remember Micro-interactions (scrolling, bookmarking, navigation) break flow
The core issue: Touchpads behave like mice, not surfaces.
Core Idea
Turn the touchpad into a spatial language.
Instead of:
“3-finger swipe = something”
Use:
Where you touch + how you move = meaning
Example:
Right edge + swipe up → Scroll Bottom edge + swipe right → Next page Top edge + hold → Enter link mode
This creates a composable, learnable system instead of a list of gestures.
Design Principles
Spatial consistency → same location = same meaning Zones + direction = actions → small system, large vocabulary Commit on release → reduces mistakes Asymmetric effort → harder to trigger than to exit Forgiving interaction → system adapts to human imprecision
System
Built on Linux using raw touch input.
Pipeline:
Capture absolute touch data (evdev) Classify → zone + direction Map → semantic action Output → simulated keypress Web reader responds
Everything is declarative — gestures can be remapped without changing code.
Interaction Design
(need a table here)
Prototype
Python-based gesture engine Real-time visualizer for debugging Web document reader controlled entirely via touchpad No cursor interaction at any point
Evaluation
Qualitative study with users performing reading tasks.
What worked
Rapid learnability (minutes, not hours) Users formed a spatial mental model, not memorization Gestures felt logical and meaningful
What broke
Hard zone boundaries → accidental triggers Corner zones → highest error rate
Key Insight
When gestures are spatially motivated, users don’t memorize — they infer.
That changes everything
Limitations
Linux-only (raw input access constraint) Small, qualitative study Controlled reading environment
Further Improvements
Add subtle feedback (haptic/audio) Track quantitative metrics (error rate, speed) Extend to multi-finger spatial inputs