RL training environments for the Microduck biped robot, with actuator modeling and sim-to-real recipes exporting deployable ONNX policies.