Microduck RL

Microduck RL provides reinforcement-learning environments for training behaviors for Pollen Robotics' small bipedal Microduck robot. It is built on mjlab with MuJoCo Warp and trains PPO policies at a 50 Hz control rate. The repository includes tasks for walking, standing up, sitting, kicking, rolling, skating, crouching, and other physical behaviors. Its sim-to-real setup models actuator physics, backlash, domain randomization, and reward design. Trained policies can be exported to ONNX and deployed through the separate Microduck runtime. Training normally uses CUDA hardware, while supported workflows can offload jobs to Hugging Face Jobs when a local GPU is unavailable.