Humanoid Robot for Education and Personal Assistance

I’m working on rebuilding the Poppy Humanoid Robot to continue their open-source legacy, making it lighter, stable, faster, cheaper and smarter using offline agentic A.I. models.

My motivation for rebuilding Poppy Humanoid as an affordable, open-source robot goes beyond just making humanoid robotics accessible. It stems from my deep disappointment with the Romanian high school education system, which failed to nurture real innovation, practical skills, and critical thinking.

⁹

Growing up, I saw firsthand how the system prioritized outdated theory over hands-on learning, discouraging students from experimenting and thinking outside the box. Robotics, AI, and engineering were treated as abstract concepts, with little to no practical application in classrooms. Many talented students were left without guidance, access to resources, or encouragement to explore technology in a meaningful way.

This project is my way of breaking that cycle—by giving anyone, regardless of background, the tools to learn, experiment, and build without being limited by expensive proprietary systems.

Technology should be open, affordable, and accessible, allowing students, makers, and researchers to develop real-world skills and push the boundaries of what’s possible.

What Will Donations Be Used For?

Your support will help cover:

• Research & development (hardware + AI)
• 3D printing, testing of new parts and new materials
• Open-source documentation & tutorials
• Creating an accessible DIY kit for everyone

This is a community-driven project, and all progress will be open-source, so anyone can contribute and benefit.

The Brain: Affordable, Offline, Child-like

The key innovation in my Poppy Humanoid rebuild is its offline brain, which will make it independent, fast, and secure while keeping it affordable.

• Hardware: A small Raspberry Pi will power the robot, keeping costs low and accessibility high.
• AI Model: A highly distilled version of GPT-NeoX, optimized to run on limited hardware without needing cloud processing.
• Capabilities: It will handle speech recognition, decision-making, and vision, making the robot functional without expensive external computing.

This offline approach ensures full privacy, making it ideal for personal and educational use.

AI Motion Compensation & HerkuleX Servos – A Smarter, Cheaper Approach

To make the redesigned Poppy Humanoid more affordable and efficient, I’m replacing the original servo motors with HerkuleX servos, which offer better performance-to-cost ratio and built-in feedback control.

Why HerkuleX?

• Cheaper than Dynamixel while offering similar features
• Daisy-chain connectivity, reducing wiring complexity
• Built-in PID control & position feedback, improving precision
• More torque per dollar, making movements stronger and smoother

AI Motion Compensation

To further optimize movement, I’m integrating an AI-driven compensation system powered by the distilled GPT-NeoX model for:

• Predictive movement adjustments – AI corrects small errors in real-time
• Energy-efficient motion planning – reducing unnecessary strain on motors
• Self-learning – The system fine-tunes movements over time

By combining HerkuleX servos with AI motion compensation, I can achieve high-quality movement at a lower cost, making humanoid robots more accessible to everyone.

Future Acrobat Robots & Motion Control

Beyond just a humanoid form, I also plan to develop an acrobat-style humanoid robot, focusing on:

• Dynamic movement – balance, flips, and agility.
• Efficient control systems – using reinforcement learning to improve performance over time.

Estimated Technical Data Sheet

This estimated tech sheet will be refined as testing progresses. The focus is on cost reduction, printability, and AI-driven autonomy, making it accessible for makers, educators, and roboticists worldwide.

• Structure & Build: FDM 3D-printed (PLA, PETG, or ABS), modular design
• Height: ~80-90 cm (adjustable)
• Weight: ~3-4 kg
• Degrees of Freedom (DoF): ~20-25 (subject to final servo selection)
• Actuators: HerkuleX servos (cheaper alternative to Dynamixel)
• Main Processor: Raspberry Pi 4 / 5 (final version TBD)
• AI Model: Distilled GPT-NeoX (optimized for offline use)
• Motion Compensation: AI-based predictive motion control
• Perception: YOLO for object recognition, LiDAR for 3D mapping, VGA camera for high-speed rendering, audio speaker and stereo microphones.
• Control Interface: Gesture recognition, voice commands, or remote control
• Connectivity: Wi-Fi, Bluetooth (optional)
• Battery: Li-ion or LiPo (estimated 12V, 5000mAh)
• Software Framework: Python-based, using PyTorch/TensorFlow for AI
• Operating System: Raspberry Pi OS (Debian-based)
• Expandability: USB, GPIO, and I²C for additional sensors and modules
• Applications: Education, research, robotics development, acrobatics

This design prioritizes affordability, accessibility, and AI-driven autonomy, making it easy to build, modify, and expand for various robotics applications.

By donating, you’re supporting the development of affordable, intelligent, and offline AI robotics for everyone.

Any donation helps make this vision a reality.

Thank you for your support!