NVIDIA Microduck: What It Is and Where to Buy (2026)

NVIDIA Microduck is not an official NVIDIA product name. The robot is Microduck, made by Pollen Robotics - the robotics team at Hugging Face - and sold as a $399 pre-order at microduck.homes. NVIDIA GPUs belong to the recommended policy-training workflow; the robot itself uses an RK3566 onboard computer.

If you found this page after searching for "NVIDIA Microduck price", "NVIDIA Microduck where to buy" or "NVIDIA Microduck AI robots", the short buying answer is:

• Want the real robot? Pre-order Microduck directly from its maker.
• Want to train policies? Use suitable CUDA hardware or a hosted Hugging Face Jobs workflow; the robot does not contain an NVIDIA GPU.
• Want to avoid look-alike stores? Use the official-source checklist before paying.

Is Microduck an NVIDIA robot?

No. Microduck is a Pollen Robotics product, and NVIDIA does not appear in its official product name. The confusion is understandable because Microduck's reinforcement-learning workflow is compute-heavy and commonly trained on NVIDIA GPUs.

The distinction matters when checking a listing: "NVIDIA Microduck" describes a search phrase, not a second robot model or an NVIDIA-branded edition. A listing that invents an NVIDIA edition, an exclusive chip variant or a different manufacturer should be treated cautiously.

What is the official Microduck robot?

Microduck is a 25 cm biped duck robot that arrives assembled, calibrated and ready to drive. It weighs under 800 g and carries 15 motors, a camera, LiDAR, two IMUs, microphones, a speaker and NFC.

The appeal is not simply that it walks. Seven trained behaviours ship installed, and the software path lets owners inspect, train and deploy new policies. The public stack includes a Python SDK, browser simulator and reinforcement-learning environments. The complete mechanical and electronic design files are not published, so Microduck is open-source software on a finished product - not open-source hardware.

For the complete technical picture, see the Microduck specs page, software architecture and GitHub guide.

Why does NVIDIA come up in Microduck training?

Training a walking policy is much more demanding than driving the finished robot. The documented local workflow calls for Linux x86_64, Python 3.12 and a CUDA-capable NVIDIA GPU with roughly 8 GB of memory or more as a practical starting point.

That GPU runs simulation and policy training. Once a policy is exported, the walking behaviour runs onboard the robot. You do not need an NVIDIA workstation beside the duck to make it stand, walk, turn, kick or get back up.

The Microduck training-compute guide explains the split between CPU, GPU and hosted training. If you already have an NVIDIA GPU, it can be useful. If you do not, Hugging Face Jobs is the documented way to avoid buying one just to try training.

How much does Microduck cost?

The robot is $399 before taxes and shipping. Optional packs are $39 for Charger, $39 for Accessory and $119 for Dev. The robot is a one-time purchase, not a subscription.

GPU hardware is not included and is not required for basic play. A training-capable GPU or hosted compute is a separate cost only when you decide to train policies. The full cost breakdown separates the robot, packs, checkout charges and optional compute.

Where can you buy Microduck?

Buy Microduck direct from microduck.homes. The site is the official pre-order route described by the maker; generic marketplace listings that reuse the name are not the same product.

Before ordering:

1. Confirm that the page identifies Pollen Robotics and the 25 cm Microduck robot.
2. Check that taxes and shipping are shown in the official checkout rather than hidden behind a marketplace listing.
3. Use Microduck support for order questions instead of sending payment details to another store.
4. Read the buyer's guide if you are unsure whether a retrainable biped fits your experience level.

Is there an NVIDIA Microduck for sale?

No NVIDIA-branded Microduck edition is published. If a store advertises one, ask who manufactures it, which physical robot you will receive and which software stack it runs before paying.

The verified buy path is the same for every retail colourway: choose Microduck on the pre-order page, select any optional pack and complete the direct checkout. NVIDIA GPUs are an optional training tool, not a retail variant of the robot.

NVIDIA Microduck FAQ

Is Microduck made by NVIDIA? No. Microduck is made by Pollen Robotics.

Does Microduck contain an NVIDIA GPU? The published onboard compute is an RK3566. NVIDIA GPUs are relevant to local policy training, not the shipped robot's control loop.

How much is an NVIDIA Microduck? There is no separate NVIDIA edition. The official Microduck is $399 before taxes and shipping.

Can I train Microduck without an NVIDIA GPU? Yes. Hugging Face Jobs is the documented hosted alternative.

Does training on NVIDIA hardware make the robot itself NVIDIA-powered? No. Training and onboard inference are separate stages with different hardware.

Where is the safe place to pre-order? Directly at microduck.homes.

Key takeaways

• "NVIDIA Microduck" is a commonly used search phrase, not an official product line.
• The official robot is Pollen Robotics' $399 Microduck.
• Its onboard computer is an RK3566; NVIDIA GPUs are an optional training resource.
• Use CUDA compute locally or Hugging Face Jobs in the cloud when training new policies.
• Pre-order only through the official direct store and verify the seller before paying.

Where to go next

• Microduck training compute - choose local GPU or hosted training
• How to train Microduck - the full sim-to-real workflow
• Microduck specs - every published hardware number
• Buying safely - identify the official product and store
• Pre-order Microduck - $399 before taxes and shipping