Nvidia's $399 Robot Duck Learns to Walk Using AI
This cheap robot could teach itself to walk—and that's a big deal for the future of robotics.
Nvidia and Hugging Face published a walkthrough showing how developers can train Microduck, a $399 duck-shaped robot, to walk in a virtual world before trying those moves in real life. The open-source, two-legged robot was developed by Pollen Robotics and Hugging Face for students, researchers, and developers experimenting with physical AI.
The walkthrough uses Nvidia’s Isaac Lab and Newton simulation tools to train walking, handling uneven ground, and recovering after falling over. The training software is available now, while the physical robot is still taking preorders. Microduck offers a smaller, less expensive way to explore physical AI.
- Microduck is a $399 robot duck that learns to walk through AI simulation, making robotics experimentation affordable for students and developers.
- The training method, sim-to-real, lets robots practice in a virtual world before trying moves in real life, which is safer and cheaper.
- The software is available now, but the physical robot is on preorder, so you can start experimenting in simulation today.
Why It Matters
This makes robot experimentation cheap and accessible, potentially speeding up innovation in robotics and AI.