ODIE V2: “Learning to Walk”

ODIE v1 proved the concept worked, but I knew the gait wasn't optimal. I couldn't see what the legs were actually doing in real time, couldn't predict failures before they happened. So for v2, I added simulation. Same hardware—8 servos, Pi Zero 2W—but now I built a Gazebo simulation and ROS2 stack to model the robot digitally. I could test gait algorithms before running them on real hardware, and inverse kinematics validation meant I could predict how the legs would move instead of just guessing. The simulation became my testing ground.

The simulation let me refine everything systematically—better gait timing, smoother transitions, less wasted motion. I could run 100 tests in simulation before touching the physical robot. Then I tested those refined algorithms on the actual hardware and they worked. The simulation matched reality. This was the bridge between raw iteration (v1) and learned optimization (v3). Simulation taught me how to think about walking as a system—coordinating legs, timing movements, minimizing energy waste—instead of just tweaking servo angles until something worked.