Pollen Robotics (Hugging Face) Microduck (pollen-robotics.com)

Pollen Robotics has launched the Microduck, a 25 cm open-source bipedal robot that users can train themselves using reinforcement learning. This innovative robot is designed to operate out-of-the-box, allowing individuals to teach it various movements and behaviors through simulation training. The Microduck uses a “sim2real” model, which means behaviors are first learned in a physics simulation and then deployed to the actual robot, facilitating a seamless transition from virtual training to real-world application. Its SDK and training framework are available on GitHub, promoting community engagement and collaboration. The significance of the Microduck for the AI/ML community lies in its accessibility and versatility, as it combines hardware and software that can be easily modified and retrained by users. It features various pre-trained policies for actions like walking, sitting, and even a unique roller-skating locomotion. This flexibility encourages user participation in advancement and innovation within the robotics space. Priced at $399, the Microduck will ship in four customizable colors, and it not only empowers individuals to engage in hands-on learning but also fosters a vibrant community on platforms like Discord, where users can exchange insights and improvements.