I know building a large quadruped with servos is not exactly the smartest engineering choice 😅 But that was the point: learning, testing, failing, improving. I designed a walking animation by hand in Godot, then used PPO-based RL to optimize the movement. Everything I learned from this will flow into my next robot project.

I just made my 3D printed quadruped robot open source! It is a servo-driven robot dog project with 3D printed parts, calibration tools, animation support, and a web-based control interface. The project is still experimental, but I wanted to share it with the maker community so others can build it, remix it, improve the mechanics, and experiment with their own walking gaits. You will need proper servo calibration, a strong power setup, and some patience.

This spider robot taught itself to stand using reinforcement learning, no pre-programmed movements. A neural network trained in a Godot physics simulation across 50 parallel instances on a server, failing millions of times until it figured out how to balance on four legs. The trained brain runs as a tiny ONNX model on an Arduino UNO Q, reading an IMU for balance and controlling 12 servos, fully autonomous, no PC needed.