Wall-E
From Amazon kit to autonomous AI robot
Building an AI Robot
What happens when Rag gets a robot kit and access to Claude and Gemini with zero robotics experience.
The spark
China's humanoid robots doing kung fu.
I saw this and thought: robots aren't science fiction anymore. They're products you can buy.
$15,000 on AliExpress
The beginning
My wife got me a robot kit off Amazon.
A Wall-E-inspired robot kit. Cute little treads, a couple of servos, an Arduino. Just a fun toy, really.
But it immediately felt like it could be so much more.
Last time I used Arduino was 15 years ago.
All I managed to do was make an LED blink.
Zero robotics knowledge. Zero electronics experience. Just Rag with an idea and some AI.
The secret weapon
Thank god for Claude and Gemini.
Without AI assistants, this project would have been impossible. They helped me learn electronics, debug serial communication, write Arduino firmware, build Python servers, train ML models, and understand things I'd never touched before.
Act 01
The Build
PS5 controller.
Drive it like a game.
- First goal: make sure I can control it with my PS5 DualSense controller
- Bluetooth pairing, joystick mapping, differential drive with -255 to +255 motor PWM
- Analog sticks for driving, triggers for speed, buttons for servos and camera
The first time it moved with the controller was pure magic.
PS5 DualSense Control
But I wanted more.
Remote control was fun. But what if it could see? Hear? Talk? Think? Navigate on its own?
Act 02
Adding Intelligence
"Yo Wall-E"
Trained a local wake word model using Picovoice's Porcupine. Always listening, zero cloud dependency. Say "Yo Wall-E" and it snaps to attention.
The full voice pipeline: wake word, transcription, LLM thinking, text-to-speech response, movement choreography.
It can see. It can recognize you.
Object Detection
YOLOv6/v8 on OAK-D
80 COCO classes at ~15 FPS. Runs on the Myriad X VPU. No Pi CPU needed for inference.
Face Recognition
Haar + HOG re-ID
Detects faces, re-identifies people with persistent memory. It remembers who you are.
Emotion Detection
FERPlus ONNX
8 emotions: happy, sad, angry, surprise, fear and more. Reacts differently to your mood.
Plus stereo depth from the OAK-D cameras for spatial awareness.
Let it roam. Let it narrate.
- Autonomous exploration mode with ultrasonic obstacle avoidance
- Environment scanning with the camera gimbal, objects get described by the LLM
- Wall-E narrates what it sees in its chosen personality
- 4 personalities: Orus the explorer, Attenborough the narrator, C-3PO the protocol droid, Wild the chaos bot
Autonomous Expedition
A full web dashboard.
- HTTPS web app at port 8080 with live camera feed, controls, and telemetry
- 35+ REST endpoints + WebSocket for real-time control
- Switch LLM provider, TTS voice, personality, and drive mode at runtime
- Works from any device on the network: phone, tablet, laptop
Web Dashboard
It has feelings. Sort of.
Wall-E's LED matrix face shows animations and emotions. When it's happy, it does a little dance. When it responds to voice, it choreographs head movements, eye animations, and servo gestures to match what it's saying.
The 16x8 pixel display is tiny but surprisingly expressive.
Voice + Movement Choreography
Act 03
Under the Hood
The hardware.
Brain
Motor Control
Eyes
Controller
Motors
Servos
Face
Sensors
Runs locally. No internet needed.
6 models running on-device, from speech recognition to object detection.
Speech-to-Text
Whisper tiny (int8)
Pi CPU
Text-to-Speech
Piper (lessac-medium)
Pi CPU
Object Detection
YOLOv6/v8 nano
OAK-D Myriad X VPU
Emotion Classification
FERPlus ONNX
Pi CPU (ONNX Runtime)
Wake Word
Porcupine
Pi CPU
Face Detection
OpenCV Haar cascade
Pi CPU
Architecture.
Cloud services (optional)
Google Gemini
Default LLM brain, TTS (8 voices), vision analysis
Anthropic Claude
Alternative LLM brain, drop-in swap
ElevenLabs
Alternative TTS, 9 voices, multilingual
Picovoice
Wake word model licensing
LLM and TTS providers are switchable at runtime via the dashboard.
From Kit to AI Companion
Act 04
What I Learned
So much to learn about the physical world.
Voltage & Amperes
What powers what. Why motors need separate power. What happens when you don't respect current limits.
Batteries & Power
Li-ion vs LiPo. Discharge rates. Voltage regulators. Why your servo jitters when the motor spins up.
Soldering & Cables
Crimping connectors. Dupont wires. Heat shrink tubing. The satisfaction of a clean solder joint.
Software is forgiving. Hardware is not. If you wire it wrong, it literally smokes.
This would never have been possible without Claude and Gemini.
Arduino firmware
Python async server
ML pipelines
Electronics debugging
The AI didn't build it for me. But it gave me the knowledge to build it myself. It was like having a patient expert sitting next to me 24/7, answering every stupid question without judgment.
The barrier to learning anything has collapsed.
Someone with zero engineering background built an autonomous, voice-controlled, object-detecting AI robot. Not because the tools are easy. Because AI made the learning curve manageable.
What now
Give it a go.
- Start small. Get a kit. An Arduino. A Raspberry Pi. The hardware is cheap. The AI assistance is free.
- Don't be afraid. I hadn't touched electronics in 15 years. Claude and Gemini walked me through every step. Every confusing datasheet. Every fried component.
- Build something physical. Software is invisible. Robots move, speak, see. There's something deeply satisfying about code that makes things happen in the real world.
Robots are the future.
And with AI, you don't need a PhD to build one.
Wall-E
Thomas Ragger
rag@wild.as
wild.as
@wild.as