suki.ENGINEERING
& CURIOSITY

INDEPENDENT PROJECT / ASSISTIVE TECHNOLOGY

A look.
A voice.
A way forward.

An eye-tracking wheelchair interface that turns a glance into an intention—and asks before it moves.

Experience the idea ↗Explore the engineering ↓
S

A project by Suki
System design → hardware → software integration

Eye-tracking wheelchair prototype with an inset close-up of the wearable glasses and camera electronics.
PROJECTGazeLink
MY ROLEEnd-to-end development
DISCIPLINESEmbedded systems · Hardware · Interaction
STAGE Prototype in development
01 / THE QUESTIONA HUMAN STARTING POINT

What if movement
could begin with
a glance?

A joystick assumes that the person using it can reliably move a hand. I wanted to explore a different starting point.

GazeLink is my exploration of an eye-controlled wheelchair interface. A camera observes the user's eyes, an ESP32-S3 interprets the direction, and a voice module asks for confirmation before the system issues a motion command.

The design question is as much about intention as it is about sensing: looking somewhere should not automatically mean moving there.

THE CENTRAL DESIGN DECISIONGaze suggests.
Voice confirms.
02 / THE INTERACTIONWATCH THE CONTROL LOGIC

One intention.
Two deliberate steps.

Watch the eye move through four directions.
A five-second dwell leads to a voice request,
a simulated reply, and a motion command.

AUTOMATIC WALKTHROUGH
↑Forward
←Left
→Right
↓Reverse

Look → dwell → voice request
→ simulated confirmation → move

READY TO OBSERVESIMULATION / NOT LIVE HARDWARE
5.0s

Looking left.

An observation, not yet a command.

↗ Eye direction maps to forward, reverse, left and right.

↗ Voice requests and replies are simulated automatically.

↗ Five seconds is the example dwell, not a measured optimum.

03 / SYSTEM DESIGNCONTROL · CONNECTIONS · POWER

System architecture

Gaze and voice control, independent fall detection, and two DC-DC converters powered by the 36 V wheelchair battery.

System architecture — gaze control, voice confirmation, fall alerts and power distribution.
System architecture — gaze control, voice confirmation, fall alerts and power distribution.

Hardware wiring

Connections between the main controller, expansion board, voice module, motor driver and sensors.

Hardware wiring — controllers, sensors, modules and power connections.
Hardware wiring — controllers, sensors, modules and power connections.
04 / FROM SCHEMATIC TO OBJECTDESIGNED & INTEGRATED BY SUKI

Models & circuit boards

Custom circuit boards, a camera mounted on glasses, and a physical assembly that brings the subsystems together.

01 / MECHANICAL DESIGN

Wearable model

The camera and main board follow the glasses frame. All three CAD views are shown below.

Front view
Front view
Rear view
Rear view
Side view
Side view
02 / ELECTRICAL DESIGN

Circuit schematics

Main PCB — ESP32-S3 controller and interfaces
Main PCB — ESP32-S3 controller and interfaces
Camera — FPC interface and supply rails
Camera — FPC interface and supply rails
Power — USB-C input and 3.3 V regulator
Power — USB-C input and 3.3 V regulator
Expansion PCB — module sockets and control signals
Expansion PCB — module sockets and control signals
03 / VISION & CONTROL

Main PCB

A narrow ESP32-S3R8 board integrates the camera interface, processing and control connections. Board outline: 70 × 14 mm.

Main PCB — front render
Main PCB — front render
Main PCB — back render
Main PCB — back render
Main PCB — layout
Main PCB — layout
Main PCB — routing detail
Main PCB — routing detail
04 / CONNECTIONS & VOICE

Expansion PCB

Connections for the ASR PRO 2.0 voice module, UART communication, motor-control signals and 5 V / 3.3 V power.

Expansion PCB — perspective render
Expansion PCB — perspective render
Expansion PCB — front render
Expansion PCB — front render
Expansion PCB — layout
Expansion PCB — layout
DECISION 01

Fit the board to the body.

The narrow main PCB follows the glasses frame. Packaging is a design constraint, not an afterthought.

DECISION 02

Separate intention from motion.

A dwell creates a request. Voice confirmation creates the command. This makes the decision explicit.

DECISION 03

Work within the pin budget.

An additional ESP32 takes on the MPU6050, red indicator and notification path when main-controller I/O is limited.

05 / AT THE WORKBENCHREAL PARTS. REAL CONNECTIONS.

Beyond
the clean diagram.

The work moves between CAD, circuit design and the bench. These photographs document the physical prototype and its assembly.

01 / PHYSICAL INTEGRATIONThe interface leaves the screen.
02 / ASSEMBLYConnecting the separate systems.
03 / COMPONENTSWorking through the physical details.
04 / WEARABLE INTEGRATIONThe glasses and wheelchair together.
05 / REAR CONNECTIONSElectronics mounted behind the chair.
06 / REFLECTION & NEXT STEPSA PROTOTYPE IS A QUESTION, TOO.

Building it is
only the beginning.

The next challenge is to understand how the interface behaves outside an ideal interaction.

This project brings together a wearable, custom electronics and a proposed control sequence. The materials shown document design and assembly; they do not yet establish reliability in everyday wheelchair use.

S
Suki

System concept, circuit and PCB design,
mechanical integration and software integration.

WHAT I WANT TO MEASURE NEXT
01 Intent recognition ＋

Compare deliberate gazes with natural eye movements across lighting conditions. Measure false requests and missed directions.

02 Dwell & confirmation ＋

Compare dwell durations and voice-recognition performance. Record task time, cancellations and confirmations before choosing a threshold.

03 Fall-alert reliability ＋

Separate ordinary chair motion from simulated fall events. Check LED behavior, message delivery and network-loss conditions.

04 Fit & control behavior ＋

Evaluate camera placement and repeatable calibration. Define and verify stop behavior and fault handling on a controlled bench setup before any occupied-chair trial.

GAZELINK / PROJECT NOTEBOOK

Download ↗