Across the global technology and industrial sectors, a small robot yet to enter large-scale delivery has become a phenomenal product. The Microduck (colloquially known as Robot Duck) bipedal robot has stirred up the global developer community, hitting a peak pre-order rate of one unit sold every four seconds. It has also rapidly raised industry attention for edge AI chips. This compact intelligent robot is far more than a viral hardware device; it has reshaped the commercialization logic of global embodied AI and delivered a new paradigm for mass production and accessible development of small-scale intelligent robots.
On August 27, Pollen Robotics, under the Hugging Face umbrella, officially launched the Microduck bipedal robot and opened global pre-orders. Standing 25 cm tall and weighing under 800 grams with a streamlined duck-like appearance, it is priced at $399 for pre-sale, breaking the long-standing niche, high-cost status of traditional bipedal robots. Market response exceeded all projections. The peak pre-order rate reached one unit every four seconds on the first day of pre-sales. Within five days of launch, cumulative global pre-orders surpassed 10,000 units, exceeding the total annual sales of the brand’s prior-generation Reachy Mini robot. The original Christmas delivery schedule was disrupted, and the brand implemented a first-order, first-ship queuing system, extending lead times for new orders to several months.
Unlike consumer service robots built for household use and large humanoid prototypes costing thousands to tens of thousands of US dollars, the Microduck has a sharply defined positioning: a software development platform for embodied AI targeting global developers, university research teams and AI enthusiasts. This core value proposition underpins its rapid popularity among the global maker and tech community.
On the hardware side, the Microduck is equipped with the RK3566 general-purpose edge AI SoC as its main controller. Its hardware suite includes 15 servo motors, a wide-angle onboard camera, laser rangefinder, dual IMU inertial measurement units, integrated microphones and speakers, alongside Wi-Fi, Bluetooth and NFC connectivity. This hardware setup fully supports motion control, environmental perception and local AI inference for small robots. Seven pre-trained behavioural policies come pre-installed out of the box, enabling autonomous walking, posture switching, duck-bill gripping, fall self-recovery and wheeled expansion; users can directly control the robot via a hand controller after unboxing.
Software ecosystem constitutes Microduck’s core competitive advantage. Its SDK, MuJoCo physics simulation environment and reinforcement learning training pipeline are open-sourced under the Apache-2.0 license. Mechanical drawings, PCB schematics and hardware BOM files remain closed-source. Developers may train robot gait, obstacle avoidance logic and interaction strategies within the simulation environment, then deploy trained AI behavioural policies onto physical hardware. Trained model weights can be uploaded, shared and downloaded across the open-source community. Developers no longer need to build algorithmic frameworks from scratch, substantially lowering the barrier to physical experimentation and iteration for embodied AI algorithms.
The phenomenal product hype has also rippled through global capital markets and lifted industry interest in edge AI chips and small intelligent robot segments. The adoption of this general-purpose edge main controller validates its suitability for small bipedal robot deployments, directing broader industry focus toward lightweight, low-cost, mass-producible embodied AI solutions.
For a long time, the humanoid robot industry has heavily invested in large-format devices, which suffer from high hardware costs and mass-production challenges. Most such products remain confined to laboratory prototypes and demo showcases, with limited pathways to large-scale commercialization. Microduck pioneers an alternative route: lightweight hardware, accessible pricing and open software ecosystems, bringing bipedal robot development within reach of individual developers and research teams.
The RK3566 is not a custom-built chip exclusively designed for robots. It is a mature, mass-produced general-purpose edge AI SoC widely deployed in tablets, edge gateways and educational hardware. A single chip handles Microduck’s dynamic motion control, machine vision recognition and real-time local AI inference, verifying that general-purpose edge chips meet the computational requirements of small bipedal robots, desktop research robots and educational robots. Compared with large humanoid robots with lengthy commercialization timelines, lightweight small robots represent a more realistic growth market for robotics and AI hardware.
The industry needs to rationally distinguish product hype from commercial delivery timelines. Microduck is still in pre-order status. Large-volume shipments, long-term equipment reliability and sustained performance in real-world scenarios remain to be validated by the market. Short-term capital market enthusiasm is driven by technological momentum and does not signal the arrival of full-scale commercialization for embodied AI. Meanwhile, small bipedal robots still face multiple technical hurdles. Beyond the main controller, continuous iteration is required for servo motor precision, sensor calibration, motion control algorithms and mechanical structural optimization.
Nevertheless, the breakout success of Microduck carries landmark industry significance. As a mass-producible robot development platform released by a leading global open-source AI community, it leverages a general-purpose edge computing backbone and open software ecosystem to connect hardware mass production, algorithm iteration and global co-development by the developer community. It pushes embodied AI out of laboratories and into a broader base of developers.
Amid the industrial wave of physical AI and embodied AI, Microduck and similar compact, mass-manufacturable robots with open software ecosystems stand out as key breakthrough vectors for the robotics industry, compared with conceptual large humanoid robots. Going forward, realising physical AI industry dividends hinges on on-site deployment of small robots, co-building of open-source ecosystems and adaptation of edge computing capacity.
