Creative Coding
Train a Robot to Balance in Simulation, Then Ship It to Real Hardware
Shawn Hymel's reinforcement-learning build trains a two-wheeled balance bot entirely in simulation — MuJoCo, Gymnasium, CleanRL's PPO trainer — then exports the trained policy to an ESP32 with no retraining on the physical robot. The repo and a six-part video series document every step, including the curriculum that makes the sim-to-real jump work.