PO-F1
PO-F1 AI Camera Pet Feeder
Wide-angle AI camera feeder with on-device multi-pet recognition, on embedded Linux with a 1 TOPS NPU — a base for custom development
A camera pet feeder reference design: wide-angle pan-tilt view, on-device pet tracking and multi-pet recognition on embedded Linux with a 1 TOPS NPU, night vision, two-way audio and offline-capable feeding schedules, with app and cloud. Adapted to your brand through a paid development phase (NRE). Not sold off the shelf.
Why it works
A wide-angle view that follows the pet
The wide-angle camera pans 360° horizontally and tilts 140° vertically, and tracks the pet automatically. It is built to find pets, not to identify people.
Knows which pet ate what
Multi-pet recognition links each meal to the right pet profile, and the log shows who ate, when and how often. Several pets, one feeder, no shared-bowl guessing.
AI that watches so the owner does not have to
It checks the bowl before dispensing, raises an alert if a pet has not eaten or moved for a long time, sorts the day into a smart timeline of highlights, and follows behaviour trends to catch early changes.
Keeps feeding when the network or the power fails
Feeding schedules keep running without Wi-Fi, and battery backup carries meals through a power cut. Dual-band Wi-Fi keeps the camera connected in a crowded home network.
Made for real kibble
Handles kibble up to 12 mm and freeze-dried pieces up to 9 mm, with multi-layer clog protection, a 5 L hopper and a detachable stainless-steel bowl that goes in the sink. Dry food only, by design.
Built for sharing
Live view around the clock, two-way audio, and up to 20 shared accounts, so the whole household can check in.
Branded app and cloud
Every project includes an owner app under your name — your logo, your colours, published from your developer account — and the cloud behind it.
Over-the-air updates
Firmware updates reach devices already in homes, rolled out in small batches first so a faulty build never reaches the whole fleet.
Edge AI platform
Right-sized edge AI: embedded Linux and a 1 TOPS NPU, chosen for the job
The feeder runs embedded Linux with a 1 TOPS NPU, sized to run pet tracking and recognition on the device and no more. That keeps the bill of materials and the cloud cost low, works without a network, and leaves room for the models to improve after launch.
- 1 TOPS
- on-device NPU for vision models
- Linux
- embedded OS, shipped in production
- On-device
- pet tracking and recognition
- Multi-pet
- individual recognition at the bowl
The software stack, layer by layer
- Feeding logic
- Per-pet portions and schedules
- Hold-back for the wrong pet
- Jam and empty-hopper handling
- Per-pet records and trends
- AI models
- Pet tracking
- Individual pet recognition
- Bowl-status detection
- Inference runtime
- NPU, 1 TOPS
- Model deployment on the device
- Model updates over the air
- Embedded Linux
- Camera image pipeline
- NPU and sensor drivers
- Power management
- Secure system updates
- Hardware
- Application processor with NPU
- Wide-angle camera and pan-tilt motors
- Load sensing
- Dispensing motor drive
Perceive
- Wide-angle camera covers the room; the pan-tilt head follows the pet
- Recognises which pet is at the bowl
- Checks the bowl before dispensing
- Infrared night vision
Decide
- Separate portions and schedules per pet
- Holds back when the wrong pet is there
- Raises an alert when a pet has not eaten or moved for a long time
- Keeps feeding on schedule when the network is down
Improve
- Updates the system, the models and the firmware over the air
- Staged rollout, so a faulty build never reaches every unit
- New recognition features can arrive after launch
Why Linux in a consumer product is hard
Boot time and power
A feeder must be ready quickly and cannot burn power idling. A general-purpose Linux image needs real trimming to behave like an appliance.
Thermals in a sealed body
A camera and an NPU running together generate heat in a housing that has to stay closed around food and pets.
Drivers that last years
Camera, NPU and sensor drivers have to keep working through years of system updates, not only on the day of the first sample.
Updates that cannot brick a unit
Updating a system, a model and the firmware over the air needs rollback and staging, because a feeder that stops working is a pet that goes hungry.
Specifications
Feeding
| Food | Dry kibble up to 12 mm; freeze-dried up to 9 mm |
|---|---|
| Hopper | 5 L |
| Protection | Multi-layer clog protection, battery backup |
| Bowl | Detachable stainless steel |
| Schedules | Run offline, without Wi-Fi |
Camera and AI
| View | Wide-angle, 360° pan and 140° tilt, high definition |
|---|---|
| Night vision | Infrared |
| Audio | Two-way talk |
| AI | Pet tracking, multi-pet recognition, bowl status, activity alerts, daily highlights |
| Processor | Application processor with an on-device NPU, 1 TOPS |
| Operating system | Embedded Linux |
| Updates | Over-the-air for the system, the models and the firmware |
App and platform
| Connectivity | Dual-band Wi-Fi |
|---|---|
| Sharing | Up to 20 accounts; live view around the clock |
| Pets | Cats and small dogs |
| App | Branded app, iOS and Android; optional AI-care subscription |
What you can change
- Housing design, colours, materials and logo
- Hopper capacity and food types
- Camera or no camera; privacy features
- App branding, cloud region and packaging
Designed toward
- CE
- FCC
- RoHS
- LFGB / FDA food contact
A reference design, not stock. We build it under your brand — colourway, logo, packaging and tuning are yours.
Request a Quote
Take the PO-F1 AI Camera Pet Feeder to your brand
Tell us your target market, quantity and the certifications you need. You get a cost envelope and a schedule from an engineer, within one business day.