Bringing Characters to Life: How a Robotics Engineer Built a Wearable AI Fursuit Powered by LattePanda

by L.P.

When creativity meets engineering, magic happens. Mekhy is a robotics engineer and founder of Dragon Shore Studios in Brazil, a creative workshop that produces cosplay and wearable accessories.

Driven by his lifelong fascination with visual effects and animatronic performances, he set out to create something that blurs the line between robotics and character performance — a fully interactive, AI-powered fursuit capable of real-time emotion, voice, and environmental interaction.

At the heart of this ambitious project lies the LattePanda 3 Delta, a compact x86 single-board computer that bridges high-level computing, hardware control, and portability — all essential for wearable robotics.

Product Introduction

This project integrates LattePanda 3 Delta as the central processing unit in a full-body wearable robotic system. The board coordinates multiple subsystems, including machine vision, audio processing, and LED/motor control, through Python and C++ scripts communicating over serial and MQTT protocols.

With its ultra-thin design, Intel N5105 processor, onboard Arduino Leonardo co-processor, and Windows 10 compatibility, LattePanda provides the computing power and flexibility required for real-time sensor fusion, local AI inference, and complex peripheral management in a compact, power-efficient platform.

Hardware and Software Components

Hardware

The LattePanda board is mounted inside a laser-cut MDF “backpack” component box — a thematic piece with painted panels, fake switches, and even a QR code linking to the control app. The modular design keeps everything neatly organized while maintaining airflow for cooling.

Lattepanda 3 Delta
BIGTREETECH 5-inch HDMI LCD displays
JC-TYC-301 USB-C HDMI adapter
P2 to USB adapter
ELP OV2710 night vision camera
Boya BY-MM1 unidirectional microphone
Small 5V NeoPixel LED strips
Trendwoo X-Bass speaker
Powerbank batteries (with different specifications)
BNO055 IMU sensor (currently not used but will be in the near future)
80mm cooler fan
40mm cooler fan
Android smart tag

Software

Each part is open-sourced in his GitHub repositories , allowing others to explore or adapt the system for their own creative projects.

Mediapipe + OpenCV Python machine vision stack with custom scikit-learn model pipeline
LiveKit agent using Silero for VAD, GPT 4o mini Transcribe for STT, GPT 4.1 Nano and Elevenlabs TTS
Unity application using Live2D Cubism SDK for Unity (communication handled via websocket)
Voicemod (communication handled via websocket using their control API)
Spotify
MQTT broker running on HiveMQ Cloud
Next.js control app, which runs on Vercel
Self-hosted Telegram bot using the Telepota library
C++ script running on AVR co-processor, which controls LEDs (communication handled via UART serial)
AnyDesk

The Making Process

1.Creating the Component Box

The main computing and power modules are mounted in a custom enclosure built from laser-cut MDF wood panels.

Components such as the LattePanda 3 Delta board, cooling fans, and batteries are secured using screws and Velcro, ensuring stability and serviceability.
The internal layout was optimized for air circulation and cable management, keeping the build neat while preventing heat accumulation.
A top-mounted hook and belt system connects the box to the head of the suit, preventing it from tilting forward while helping to distribute weight evenly.
After assembly, the surface was spray-painted and decorated with stickers, faux switches, and a printed QR code linked to the control interface, enhancing both usability and thematic coherence.

2.Building the Head Mask

The head mask was 3D-printed to host the main visual and audio components, including the displays, camera, and microphone.
The inner structure leaves a central gap for facial tracking and proper microphone placement.
Two neodymium magnets embedded in the forehead area allow the mask to be easily removed for cleaning or maintenance.
The head unit integrates compact LED lighting and airflow channels, balancing visual realism with electronic safety.

3.Assembling the PCBs

Two custom-designed compact PCBs connect the LED strips, cooler fans, IMU sensor, and expansion servo ports to the LattePanda’s GPIO via a single flat cable.
The boards were designed for minimal space consumption and quick replacement, reducing wiring complexity.
A 1000 µF electrolytic capacitor was added to stabilize current flow, preventing flicker when driving long LED strips or high-load peripherals.

4.Software Deployment and Integration

The system software is divided into multiple repositories, each focusing on one key function:

Why Choose LattePanda?

LattePanda serves as the central brain of the suit, providing:

Native Windows 10 support, essential for Voicemod, Unity, and Spotify.
Powerful x86 performance to run machine vision, AI models, and audio processing simultaneously.
Onboard GPIO with AVR co-processor for LED and motor control.
Extensive documentation and community support for quick integration and troubleshooting.

LattePanda’s combination of processing power, flexible I/O, and open-source support makes it the only feasible board for a project of this scale and complexity.

Conclusion

From robotics competitions to wearable AI characters, his journey showcases what happens when imagination meets powerful embedded computing. His “DYNAMO Electronic Fursuit” isn’t just a costume — it’s a living, breathing example of creative engineering.

The LattePanda 3 Delta enabled him to unify multiple disciplines — machine vision, real-time audio, interactive graphics, and control — into a single cohesive platform that can be worn and performed live.

As he continues to refine his design and expand his project’s capabilities, one thing is clear: the future of cosplay and robotics is not pre-programmed — it’s alive, responsive, and powered by LattePanda.