Cleo
Cleo from the front
Cleo Object Detection
Cleo CAD
Cleo Wiring

Cleo — AI has a mind. We gave it a body.

Inspiration

The spark came from the Sesame Robot — an open-source quadruped that proved a walking robot could be built cheaply and accessibly. We wanted to take that foundation further.

Look at what's happening globally. China is betting big on AI hardware, humanoid robots, autonomous drones, robotic manufacturing, while North America continues to double down on software. The AI marathon isn't just being run in data centres anymore. It's being run on legs.

We believe the last frontier of AI isn't in the cloud. It's physical. It's embodied. And it starts with giving AI a body it can act through.

So we asked ourselves: what if anyone could build that?

Cleo is our answer. A low-cost, open-source, AI-powered quadruped robot built in 3 days for under $100. Designed from the ground up to be replicated, remixed, and improved by anyone with a 3D printer and a Raspberry Pi. A boilerplate for the next generation of autonomous machines, built in Canada, to keep Canada in the race.

What It Does

Meet Cleo 🐾

Cleo is a fully autonomous four-legged robot that sees, thinks, speaks, and moves. Not because she's remote controlled, not because she's pre-programmed, but because you give her a mission and she figures out the rest.

🗣️ Talk to her — speak a mission in plain English and she listens
👁️ She sees you — a front-mounted camera identifies objects and people in real time
🧠 She thinks — an onboard Gemma 4 model running locally on the Pi decides her next physical action based on what she sees — no cloud, no internet required
🚶 She moves — eight servo motors drive four legs in a coordinated gait toward her goal
😊 She feels — an OLED face displays her internal state: searching, thinking, found, complete
💃 She performs — wave, dance, celebrate when the mission is done

But what makes Cleo truly different is her modular sensor bay, a swappable hardware slot on her chassis that turns her into a specialist. Snap in a module and her entire behaviour profile changes:

Module	Mode	What she becomes
PIR motion sensor	Security	Patrols and alerts on intruders
SPO2 + vitals sensors	Medical	Emergency responder assistant
Temperature + humidity	Environmental	Hazard monitoring robot
Camera + depth sensor	Search & Rescue	Victim identification in tight spaces

Our demo scenario is search and rescue. Cleo navigates confined spaces too small or dangerous for humans, identifies victims using computer vision, and alerts first responders with location data. One robot. Swappable purpose. Infinite applications.

She could be your personal assistant, a supply chain operations robot, a security patrol unit, or the first responder that goes in before the humans do.

How We Built It

Hardware

Component	Purpose
Raspberry Pi 5	Main brain — runs all pipelines
8× MG90S metal gear servos	Locomotion — 2 DOF per leg
PCA9685 PWM driver	Servo control over I2C
USB webcam with mic	Vision + voice input
SSD1306 128×64 OLED	Expressive face display
USB speaker	Voice output
DLG 1500mAh 2S LiPo	Power source
20A buck converter	Steps 7.4V → 5V rail
HC-SR501 PIR sensor	Security module
DHT11 sensor	Environmental module
Custom 3D printed chassis	PLA+, designed from scratch

Software

Technology	Role
Python 3	Full software stack
OpenCV + MobileNet SSD	Real-time object detection
Gemma 4 (local)	Onboard LLM mission decision engine
`faster-whisper`	Local speech-to-text
gTTS / ElevenLabs	Text-to-speech
`adafruit-pca9685`	Servo I2C control
`luma.oled` + Pillow	OLED face rendering

How It All Flows

You speak a mission
    → faster-whisper transcribes locally on Pi
        → vision pipeline captures frame + detects objects
            → Gemma 4 receives mission + detections (fully local, no cloud)
                → returns single action: walk_forward / turn_left / complete
                    → PCA9685 drives servos
                        → OLED face updates
                            → loop repeats until mission complete

No cloud robotics middleware. No ROS. No co-processor. Just a Raspberry Pi, Python, and a lot of wire.

Challenges We Ran Into

Powering the Raspberry Pi from battery The Pi 5 normally draws up to 3A over USB-C with built-in protection circuitry. Powering it directly from a LiPo through a buck converter requires careful voltage regulation and protection circuits we couldn't add cleanly within the build window. Getting this right without frying a $100 computer under time pressure was one of our biggest hardware challenges.

Space allocation inside the chassis Fitting a Raspberry Pi 5, PCA9685, buck converter, LiPo, OLED, camera, speaker, and sensor bay into a chassis small enough to be a quadruped, while leaving room for eight servo cables, required multiple CAD iterations and creative stacking. Every millimetre counted.

Wire management without a PCB A custom PCB would have made this clean. Instead we sourced individual components from a local electronics shop and hand-wired everything ourselves.

Accomplishments We're Proud Of

Built a fully walking quadruped from scratch in under a week
End-to-end autonomous mission loop: voice in, physical action out, no human in the loop
Modular sensor bay that changes the robot's entire purpose with a hardware swap
Expressive OLED face that makes the robot's internal state legible to anyone watching
Entire stack on a single Raspberry Pi 5: no cloud compute, no co-processor
Under $100: easily replicable by any student or maker

What's Next for Cleo

Make her compact: redesign the chassis to be significantly smaller, closer to the Sesame Robot's footprint
Replace bulky USB peripherals: swap USB speaker, mic, and webcam for smaller embedded components (I2S microphone, CSI camera, Class D amp)
Custom PCB: replace the hand-wired harness with a proper PCB that integrates PCA9685, power regulation, and sensor headers cleanly
Proper battery protection circuit: add dedicated power management for safe direct battery powering of the Pi 5
Full open-source release: detailed build instructions, CAD files, and code so anyone anywhere can build their own Cleo in a weekend