Pollen Robotics Released Open-Source Microduck Biped
The 780-gram bipedal platform aims to bridge the gap between simulation-based AI training and physical robot hardware.
Updated on Oct. 2, 2026 in Robotics
Live Poll
Do you prefer technology designed for industrial utility over devices with personalities or playful features?
Pollen Robotics has launched Microduck, a small-scale open-source bipedal robot designed to facilitate physical AI experimentation. The platform provides a low-cost testing environment for developers to transition trained behaviors from simulations to real-world hardware.
Why it matters
The development seeks to minimize the high costs and mechanical risks associated with training physical artificial intelligence on large-scale humanoid robots. By providing an accessible hardware target, the project accelerates the refinement of locomotion and self-recovery algorithms.
Microduck stands 25 centimeters tall and operates with 15 individual motors governed by a 50 Hz onboard control loop. The hardware integrates a front-facing camera, LiDAR, and dual inertial measurement units to navigate its environment.
The players
Pollen Robotics
A developer of open-source robotics hardware and software systems focused on accessible artificial intelligence research.
The details
The system relies on an open-source software stack that hosts seven pre-trained behaviors for immediate deployment. Developers train complex movements on external machines before uploading the learned models to the hardware. A specialized self-recovery routine allows the robot to autonomously regain its standing position after falling, an essential capability for physical testing.
Timeline
October 2, 2026: The project and its specifications were officially published.
The Tech Race
Microduck extends the trend of hardware-agnostic, open-source robotics by moving the research focus from software simulation to small-scale physical embodiment. This approach follows the precedents set by the Open Source Robotics Foundation's ROS middleware to democratize hardware access for AI researchers.
Developers can access the complete software stack on GitHub and Hugging Face to begin deploying custom behaviors to their own hardware. The platform is designed for research and prototyping rather than consumer use, requiring familiarity with external machine-learning training pipelines.
The takeaway
Pollen Robotics is shifting the focus of physical AI training toward lower-cost, safer hardware benchmarks. Interested researchers should monitor the GitHub and Hugging Face repositories for future software updates and additional behavioral library expansions.
Further reading
For more developments in open-source hardware, visit Robotics.
Source note: This article includes information reported by LAFM.
Live Poll
Do you prefer technology designed for industrial utility over devices with personalities or playful features?
