Microduck robot trains in simulation before deployment

Microduck is an affordable open-source bipedal robot that can be trained to walk using simulation software before being deployed on the physical hardware. The workflow involves training a walking policy in a simulated environment with domain randomization to account for real-world factors like friction and mechanical delays, then exporting that policy as an ONNX file to run on the actual robot. This matters because it bridges the gap between simulation and real robots by using techniques that prepare policies for the unpredictability of physical hardware. The process allows developers to share trained policies on a hub, test them in a competition arena, or deploy them directly on their own robot.