Under $400: Hugging Face launches Microduck, an open-source mini bipedal robot
Pollen Robotics, an open-source robotics startup backed by Hugging Face, has released the source code for Microduck, a compact bipedal robot. Measuring just 25cm and weighing 800g, this charming machine duck combines reinforcement learning, Sim2Real transfer, and real-time control to serve as a powerful platform for miniature embodied AI research.

Compact Design, Impressive Capabilities
Powered by a Rockchip RK3566 processor, Microduck controls 15 servos at a 50Hz frequency. It demonstrates a range of practical skills, including walking via a game controller, self-righting after falls, and picking up objects with its beak. It can also sit, stand, kick a ball, and perform forward rolls. By attaching wheels to its feet, the robot switches to a different reinforcement learning strategy, enabling it to slide, turn, navigate slopes, and even stand while roller-skating.
The team humorously notes: "No ducks were harmed in the making of this robot, although we consulted several ducks."
Reinforcement Learning Powers Nearly Every Movement
The associated training project, microduck_rl, is now open-source. The technical pipeline leverages the MuJoCo / MuJoCo Warp simulation environment, applying the PPO algorithm with domain randomization to achieve Sim2Real transfer. The resulting ONNX model is deployed directly on the robot, with strategies trained at 50Hz to match the hardware's control rhythm.
To bridge the sim-to-real gap, the training environment accounts for motor friction, battery voltage fluctuations, control latency, and servo gear backlash, significantly enhancing deployment stability.
Robust Software Architecture with Service-Oriented Design
Built in Rust without heavy frameworks, the software is modularized into independent services:
robotd: Handles real-time control, reinforcement learning execution, and safety protocolsupdaterd: Manages OTA updates, signature verification, and automatic rollback on failureconfigd: Oversees Wi-Fi connectivity and device identitybtd: Manages Bluetooth communicationpadd: Supports game controller inputmediad: Handles camera, audio, and WebRTC streamingtofd: Processes ToF depth sensor dataServices communicate via Unix Socket and JSON-RPC 2.0. Only robotd holds motor control permissions. Inputs from mobile apps, controllers, or custom scripts are treated as "action intentions" and must pass safety checks in robotd before execution, ensuring architectural reliability.
Seamless OTA Updates Prevent Bricking
The update process is robust: it downloads the release, verifies the signature, switches versions, restarts services, and performs a health check. If errors occur, the system automatically reverts to the previous version, effectively preventing the device from becoming unusable.
More Than a Toy: A Miniature Embodied AI Platform
Microduck is far more than a novelty item. It integrates reinforcement learning, Sim2Real transfer, ONNX deployment, Rust-based real-time control, WebRTC media, and full OTA support into a 25cm, 800g package.
For developers and researchers focused on reinforcement learning, motion control, and embodied AI, this open-source project offers a complete pipeline from simulation to real-world deployment. With both the runtime and training code available, interested developers can explore the official repository for further experimentation.
Related article
Amazon Launches Its Own Jev Clone as Decision Models Flood the Web
Amazon Web Services has launched an open-source decision model inspired by TypeSafe’s Jev, as AI developers increasingly prioritize intelligence optimized for computer automation over frontier large language models.Released the same week OpenAI annou
Google Gemini Deeply Integrated with Google Photos for Personalized AI Image Generation
Google has recently rolled out a significant update to its AI assistant, Gemini, by deeply integrating it with Google Photos. This enhancement allows Gemini to directly access visual data from your photo library, enabling the creation of more persona
Reddit Masterpiece Makes Claude Interact with You Through Handwriting in Real Time, Providing an Immersive Learning Experience That Is Simply Outstanding
A Reddit developer has launched a new interactive handwriting app that lets users converse with Claude AI in real time. By mimicking the magical, two-way exchange of Tom Riddle’s diary from Harry Potter, the tool transforms digital note-taking into a
Related Special Topic Recommendations
Comments (0)
0/500
Pollen Robotics, an open-source robotics startup backed by Hugging Face, has released the source code for Microduck, a compact bipedal robot. Measuring just 25cm and weighing 800g, this charming machine duck combines reinforcement learning, Sim2Real transfer, and real-time control to serve as a powerful platform for miniature embodied AI research.

Compact Design, Impressive Capabilities
Powered by a Rockchip RK3566 processor, Microduck controls 15 servos at a 50Hz frequency. It demonstrates a range of practical skills, including walking via a game controller, self-righting after falls, and picking up objects with its beak. It can also sit, stand, kick a ball, and perform forward rolls. By attaching wheels to its feet, the robot switches to a different reinforcement learning strategy, enabling it to slide, turn, navigate slopes, and even stand while roller-skating.
The team humorously notes: "No ducks were harmed in the making of this robot, although we consulted several ducks."
Reinforcement Learning Powers Nearly Every Movement
The associated training project, microduck_rl, is now open-source. The technical pipeline leverages the MuJoCo / MuJoCo Warp simulation environment, applying the PPO algorithm with domain randomization to achieve Sim2Real transfer. The resulting ONNX model is deployed directly on the robot, with strategies trained at 50Hz to match the hardware's control rhythm.
To bridge the sim-to-real gap, the training environment accounts for motor friction, battery voltage fluctuations, control latency, and servo gear backlash, significantly enhancing deployment stability.
Robust Software Architecture with Service-Oriented Design
Built in Rust without heavy frameworks, the software is modularized into independent services:
robotd: Handles real-time control, reinforcement learning execution, and safety protocolsupdaterd: Manages OTA updates, signature verification, and automatic rollback on failureconfigd: Oversees Wi-Fi connectivity and device identitybtd: Manages Bluetooth communicationpadd: Supports game controller inputmediad: Handles camera, audio, and WebRTC streamingtofd: Processes ToF depth sensor dataServices communicate via Unix Socket and JSON-RPC 2.0. Only robotd holds motor control permissions. Inputs from mobile apps, controllers, or custom scripts are treated as "action intentions" and must pass safety checks in robotd before execution, ensuring architectural reliability.
Seamless OTA Updates Prevent Bricking
The update process is robust: it downloads the release, verifies the signature, switches versions, restarts services, and performs a health check. If errors occur, the system automatically reverts to the previous version, effectively preventing the device from becoming unusable.
More Than a Toy: A Miniature Embodied AI Platform
Microduck is far more than a novelty item. It integrates reinforcement learning, Sim2Real transfer, ONNX deployment, Rust-based real-time control, WebRTC media, and full OTA support into a 25cm, 800g package.
For developers and researchers focused on reinforcement learning, motion control, and embodied AI, this open-source project offers a complete pipeline from simulation to real-world deployment. With both the runtime and training code available, interested developers can explore the official repository for further experimentation.
Amazon Launches Its Own Jev Clone as Decision Models Flood the Web
Amazon Web Services has launched an open-source decision model inspired by TypeSafe’s Jev, as AI developers increasingly prioritize intelligence optimized for computer automation over frontier large language models.Released the same week OpenAI annou
Google Gemini Deeply Integrated with Google Photos for Personalized AI Image Generation
Google has recently rolled out a significant update to its AI assistant, Gemini, by deeply integrating it with Google Photos. This enhancement allows Gemini to directly access visual data from your photo library, enabling the creation of more persona
Reddit Masterpiece Makes Claude Interact with You Through Handwriting in Real Time, Providing an Immersive Learning Experience That Is Simply Outstanding
A Reddit developer has launched a new interactive handwriting app that lets users converse with Claude AI in real time. By mimicking the magical, two-way exchange of Tom Riddle’s diary from Harry Potter, the tool transforms digital note-taking into a





Home






