AI wheelchair safety add-on

CAREC helps wheelchair users notice obstacles sooner.

A clip-on safety companion for powered wheelchairs. It watches the path ahead, classifies distance locally, and gives clear sound cues before contact.

Clip-on hardwareOffline AI alertsNo wheelchair rewiring

Watch Demo See How It Works

CAREC is an open-source prototype in validation. It is not a certified medical device and does not replace caregiver supervision or professional clinical judgment.

CAREC

Forward safety monitor

Active zoneWarning

Latency

164 ms

Battery

87%

Mode

Offline

BLE

Connected

Caregiver event

Warning zone detected ahead. Slow audio cue active. No video uploaded.

CAREC concept showing a child in a powered wheelchair with obstacle awareness

Product idea

A safety layer the user can understand without looking at a screen.

What

A clip-on obstacle awareness add-on for powered wheelchairs.

Why

Indoor spaces are tight, busy, and hard to judge while moving.

How

Local AI watches the path and turns distance into simple sound cues.

The solution

Four steps from path awareness to user cue.

The user experience stays simple. The engineering behind it handles sensing, inference, zone logic, and alerts.

Camera sees ahead

Forward field of view

AI checks distance

On-device inference

Zone is selected

Clear · Warning · Critical

User hears cue

No beep · Slow beep · Fast beep

Step 1

Clip on CAREC

Mount the forward-facing sensing module without drilling or changing the powered wheelchair controls.

Step 2

Detect the path

The camera and edge AI pipeline watch for people, furniture, walls, and tight corners ahead.

Step 3

Warn the user

Simple beep patterns and display colours translate distance into clear caution or stop-now cues.

AI safety features

A practical edge-AI safety loop for real rooms.

CAREC combines local computer vision, deterministic alert zones, Bluetooth status, and open engineering documentation.

On-device AI

Obstacle detection runs locally so safety alerts still work in hospitals, basements, and offline spaces.

Three alert zones

Clear, warning, and critical zones map to silence, slow beeps, or fast beeps for quick interpretation.

Caregiver app

Bluetooth status and event logs help caregivers confirm readiness and review recent alert activity.

Privacy-first design

Core video inference stays on-device; event logs describe zones and timing, not private video streams.

Open-source build

Firmware, validation notes, and architecture docs are available for accessibility engineers to inspect.

Smart-home ready

Optional WiFi, MQTT, Home Assistant, and Node-RED paths can notify caregivers of repeated events.

Demo

See the wheelchair safety concept in motion.

The demo shows the concept as a user-facing assistive layer: the system observes the path, classifies risk, and communicates with simple feedback.

CAREC wheelchair safety concept with a child using a powered wheelchair

Care settings

Designed for homes, classrooms, therapy rooms, and clinics.

The goal is a non-invasive safety layer that can be explained quickly to users, families, therapists, doctors, and accessibility researchers.

Who it helps

Useful context for every person around the wheelchair user.

CAREC keeps the immediate alert simple while making technical status and history available to caregivers and reviewers.

Wheelchair users

Simple sound cues while navigating homes, classrooms, clinics, and therapy spaces.

Caregivers

A fast way to verify the add-on is ready and review recent warning or critical-zone events.

Therapists

A supervised training aid for distance awareness, turning practice, and indoor navigation routines.

Researchers

A reference architecture for edge AI safety loops, alert latency, and privacy-preserving event logs.

Alert zones

Three distance states. One clear cue system.

The user does not need to read a dashboard while driving. Proximity is reduced to distinct audio and visual states.

Clear

100 cm+

No beep

Warning

60-100 cm

Slow beep

Critical

0-60 cm

Fast beep

Safety boundaries

It warns. It does not take over.

The project is intentionally scoped as a passive assistive layer. That boundary matters for user trust, caregiver review, and engineering validation.

CAREC warns but never drives, brakes, or overrides the wheelchair.

Original controls and caregiver supervision remain unchanged.

Core alerts work without internet access.

No cloud video is required for the safety loop.

The clip-on mount is designed to be removable.

Current status is prototype validation, not certified medical-device deployment.

Engineering snapshot

Built as an inspectable embedded systems project.

The public project includes firmware architecture, validation goals, and implementation notes for contributors and reviewers.

Alert latency target

Under 200 ms

Runtime target

50+ hours

Core mode

Offline

Mounting

Non-invasive clip-on

Event data

Zone and timestamp

License

MIT open source

View GitHub Discuss A Similar Build

FAQ

Common questions about CAREC.

Clear boundaries make the project easier to evaluate for caregivers, clinicians, and engineers.

Does CAREC control the wheelchair?

No. CAREC is a passive warning layer. It gives audio and visual alerts but does not steer, brake, accelerate, or modify the wheelchair controls.

Does it need WiFi to work?

No for core safety alerts. WiFi is only for optional firmware updates and smart-home or caregiver notification integrations.

Is video uploaded to the cloud?

The intended default safety loop runs inference on-device. Event logs are designed around zone and timing information rather than cloud video upload.

Is CAREC ready for unsupervised use?

No. It is an open-source prototype in active validation and should be treated as a supervised assistive technology project, not a certified medical device.

Open source

Review the architecture or start a similar embedded AI build.

CAREC is one example of privacy-aware edge AI for accessibility. The same design discipline applies to sensing, alerting, OTA, and connected-device projects in other domains.

GitHub Contact