Hugging Face's Pollen Robotics Unveils $399 Open-Source Robot Duck; Nvidia Reportedly to Acquire Hugging Face
Hugging Face's robotics unit Pollen Robotics launched Microduck, a $399 open-source robot duck that walks, kicks, skates, and picks up objects. It ships with trained policies, an SDK, and a simulation environment, with training taking one to two hours on CUDA GPUs. Separately, Nvidia has reportedly agreed to acquire Hugging Face for $12.9 billion.
Pollen Robotics, the robotics team under Hugging Face, recently released a robot duck named Microduck. Standing 25 centimeters tall and weighing less than 800 grams, the robot features two short legs and a beak that opens and closes. Priced at $399, or approximately RMB 2,800, it can walk, kick a ball, and roller-skate, as well as bend down to pick up small objects with its beak. After falling, it can adjust its posture and stand up on its own.
Microduck comes with a range of pre-trained motion policies and offers an SDK, a simulation environment, and reinforcement learning tools. Developers can create large numbers of virtual robot ducks in the MuJoCo physics simulation environment for training. The official training example can run 4,096 simulation environments simultaneously; using a CUDA GPU, training a usable walking gait takes about one to two hours. The resulting neural network policy can be exported and deployed to a physical robot duck, a process known in robotics as Sim-to-Real. If the real-world performance is unsatisfactory, developers can return to the simulation environment, adjust parameters, and retrain before redeploying. Developers may train locally or submit training jobs to cloud GPUs via Hugging Face Jobs.
Hugging Face began expanding into robotics in 2024. In March 2024, Rémi Cadene, who had worked on Tesla's Optimus project, joined Hugging Face to lead open-source robotics initiatives. Two months later, LeRobot was launched, an open-source robotics learning toolkit that allows developers to share models, datasets, and training methods. According to IEEE Spectrum, the number of robot datasets on LeRobot has grown from 1,145 at the end of 2024 to more than 58,000, making it the largest dataset category on the Hugging Face Hub. Nvidia, Alibaba, and research institutions have all uploaded models to Hugging Face.
Hugging Face subsequently began developing low-cost robots, releasing the SO-100 and SO-101 robotic arms, each priced in the hundreds of dollars. In 2025, Hugging Face acquired Pollen Robotics, a company that had been developing open-source robots for nine years and whose research robot is the Reachy 2. Following the acquisition, Hugging Face launched the Reachy Mini and Microduck, both priced at $399.
He noted that building something with one's own hands is itself a joy, and expressed hope that users would continue to modify the robot after assembly, such as adding wheels, changing motors, or attaching a pair of hands.
According to The Information, Nvidia has agreed to acquire Hugging Face for $12.9 billion, a report later cited by Reuters. Nvidia was one of the investors in Hugging Face's 2023 funding round. In robotics, Nvidia's GR00T model and Isaac Lab-Arena have already been integrated into the LeRobot ecosystem. When Hugging Face and Nvidia jointly presented their Physical AI collaboration in early 2026, they disclosed that Nvidia has approximately 2 million robotics developers, while Hugging Face has more than 13 million AI developers.
Microduck's official training relies on CUDA GPUs because its reinforcement learning training uses MuJoCo Warp, and this official pipeline depends on CUDA. Two days after the release, some developers attempted to migrate the training environment to a solution supporting AMD GPUs; this version is not officially provided and has only completed partial tasks, with no verification on physical hardware yet. AMD continues to offer LeRobot-related models optimized for ROCm within the Hugging Face ecosystem.
Why this event matters
The event has a measured impact on 3 industrys. The strongest current signal is positive for Robotics, with intensity 70/100 and 80% confidence over a short term horizon.
Robotics
- Direction
- positive
- Intensity
- 70
- Confidence
- 80%
- Horizon
- Short term
Artificial Intelligence
- Direction
- positive
- Intensity
- 60
- Confidence
- 75%
- Horizon
- Short term
Semiconductor Value Chain
- Direction
- positive
- Intensity
- 50
- Confidence
- 70%
- Horizon
- Medium term
Impact figures are analytical estimates that combine direction, intensity, confidence and event importance. They are not investment advice.