{"id":114591,"date":"2026-08-30T00:53:24","date_gmt":"2026-08-30T00:53:24","guid":{"rendered":"https:\/\/youzum.net\/hugging-face-unveils-microduck-a-399-open-source-25-cm-biped-you-train-with-reinforcement-learning\/"},"modified":"2026-08-30T00:53:24","modified_gmt":"2026-08-30T00:53:24","slug":"hugging-face-unveils-microduck-a-399-open-source-25-cm-biped-you-train-with-reinforcement-learning","status":"publish","type":"post","link":"https:\/\/youzum.net\/fr\/hugging-face-unveils-microduck-a-399-open-source-25-cm-biped-you-train-with-reinforcement-learning\/","title":{"rendered":"Hugging Face Unveils Microduck: A $399 Open-Source 25 cm Biped You Train with Reinforcement Learning"},"content":{"rendered":"

Most robotics launches ask you to trust a demo video. Pollen Robotics<\/a>, the Bordeaux robotics team at Hugging Face<\/a>, is instead shipping the training loop. This week it opened pre-orders for Microduck<\/a>, a 25 cm bipedal robot in which every movement \u2014 walking, sitting, kicking, roller-skating, standing back up after a fall \u2014 is a neural policy trained in a physics simulator and exported to the hardware. It costs $399. The training environments, the reward functions, the domain-randomization settings, and the sim-to-real recipe are all public on GitHub. Microduck follows Reachy Mini, which has shipped more than 10,000 units, but reverses its premise: where Reachy Mini was built to sit on a desk and interact, Microduck is built to leave the desk, fall over, and get back up.<\/p>\n

The Hardware<\/strong><\/h2>\n

Microduck is 25 cm tall, 14 cm wide, and under 800 g. It carries 15 motors across legs, neck, and head, plus an articulated beak that picks objects off the floor. Compute is a Rockchip RK3566 with an AI accelerator, 1 GB of RAM, and 32 GB of storage.<\/p>\n

The sensor stack is unusually complete for the price. A front camera sits behind a dedicated camera-use indicator. Two IMUs are fitted, one in the body and one in the head. Range sensing is a compact LiDAR, an 8\u00d78 time-of-flight matrix. There are microphones and a speaker, two NFC antennas, plus Wi-Fi and Bluetooth. Power is a removable NP-F550 battery, 2600 mAh, good for about an hour.<\/p>\n

Seven trained moves ship in the box, driven by a bundled game controller before you write code: walk, sit and stand, kick, grab, roller-skate, and self-recovery. The robot does not speak. Each unit generates its own audio identity on first wake and keeps that voice permanently.<\/p>\n

\n<\/div>\n

\n

How the behaviors are actually trained<\/strong><\/h2>\n<\/p>

Policies are trained in microduck_rl<\/a>, built on mjlab<\/a> (MuJoCo Warp) with PPO. Pollen reports roughly one to two hours on a CUDA GPU for a usable gait at 4096 parallel environments. Without a local GPU, appending --hf-jobs<\/code> runs the same command on Hugging Face Jobs.<\/p>\n

The sim-to-real work sits in the actuator model. Each servo uses the BAM<\/a> M6 model of the Dynamixel XL330 \u2014 voltage control law, back-EMF, and Coulomb, Stribeck, and load-dependent friction \u2014 rather than an ideal PD controller. Per-environment randomization covers battery voltage, voltage sag under load, command delay, and friction magnitude. Backlash variants train against \u00b11\u00b0 of gear play, 2\u00b0 total, in series with each of the 14 servo joints in the RL layout. Because the real encoder sits on the output side of that play, the observations read through it.<\/p>\n

Trained policies export to ONNX with the observation normalizer baked into the graph. Pollen warns against deploying hand-converted checkpoints for exactly this reason.<\/p>\n

On the robot, a Rust runtime drives the 50 Hz control loop and the motor bus. Every policy shares a 61-dimensional actor observation: 48 proprioception dimensions plus commands for twist (3), head pose (4), and body pose (6). That shared contract is what lets walk, recover, and trick policies hot-swap mid-run. Environments that ignore a command slot zero-pad it rather than dropping it.<\/p>\n

The published registry covers 13 tasks: velocity tracking, stand-up, sit-stand, ground pick, ball kick (70 mm, 15 g ball, actor ball-blind), roulade, and five roller-skating environments.<\/p>\n

Key Takeaways<\/strong><\/h2>\n