Hugging Face and Pollen Robotics launched Microduck, a $399 open-source 25 cm bipedal robot and sim-to-real reinforcement learning training stack

Decision brief

What changed

Hugging Face and Pollen Robotics launched Microduck, a 25 cm bipedal robot priced at $399 and opened pre-orders, with first deliveries targeted before Christmas. The release includes an open-source software stack for sim-to-real reinforcement learning, including an SDK, simulator, and publicly available training components on GitHub.

Why it matters

This expands Hugging Face’s open AI platform strategy from models and software into low-cost robotics hardware, creating a more accessible entry point for teams that want to experiment with embodied AI and reinforcement-learning workflows. For decision-makers, the combination of low upfront cost and open tooling could reduce prototyping barriers for education, developer relations, and early robotics R&D, but it also raises practical questions about privacy, device management, and whether the hardware is robust enough for business use beyond experimentation.

Affected roles

CEO CTO CISO CMO

Evidence

All four cited pieces consistently report the core facts: Hugging Face and Pollen Robotics launched Microduck, priced it at $399, opened pre-orders, and paired it with open-source sim-to-real RL tooling. Product Hunt, TechCrunch, The New Stack, and MarkTechPost independently align on the launch, price, form factor, and software stack, while TechCrunch specifically notes privacy considerations tied to camera and microphone access.

What remains uncertain

Coverage does not establish production volume, actual delivery reliability, enterprise support terms, hardware durability, or total cost once optional accessories are included. It is also unclear from the coverage how much of the full system is open source in practice versus software-only openness, and whether the robot’s capabilities will generalize beyond demos and developer experimentation.

Monitor next

Watch for the public GitHub repositories, developer documentation, and first customer shipment confirmations to verify how complete the open training stack is and how quickly third parties can reproduce or extend behaviors.

Analytical support, not advice — assumptions and open questions stated above.

Source coverage