Google showcases Gemma 4-powered Open Duck robot that anyone can 3D print at home

Google just made building your own AI-powered robot about as accessible as assembling IKEA furniture. At Google I/O 2026, the company demonstrated the Open Duck robot, a compact, 3D-printable machine running on its latest Gemma 4 open-weight AI model, performing agentic tasks entirely on-device without needing a cloud connection.

The robot is based on the Open_Duck_Mini project, an open-source initiative hosted on GitHub that lets anyone with a 3D printer and roughly $400 in parts build their own miniature version of Disney’s BDX Droid. Google paired this hardware platform with Gemma 4 to show what local AI inference looks like when it walks around on two legs.

What Gemma 4 actually does here

Gemma 4, launched on April 2, 2026, is Google’s latest family of open-weight AI models. The model weights are freely available for anyone to download, modify, and deploy, all under a full Apache 2.0 license. That’s a first for the Gemma series.

The model family supports multimodal inputs, meaning it can process text, images, and other data types simultaneously. Context windows range from 128K to 250K tokens depending on the model variant.

Several optimized sizes ship within the Gemma 4 lineup, specifically designed to run on phones, laptops, and edge hardware.

At the I/O demo, attendees could interact with the Open Duck robot live through the Gemma Playground. Google also made the 3D-printable files available for download, essentially handing visitors the blueprint to replicate the entire project at home.

The Open_Duck_Mini project

The robot itself predates Google’s involvement. Open_Duck_Mini is a community-driven, open-source GitHub project that set out to create a miniature walking robot modeled after Disney’s BDX Droid. The bill of materials comes in under $400.

Google’s contribution was layering Gemma 4’s on-device AI capabilities onto this existing hardware. The result is a robot that can perform what Google calls “agentic tasks,” essentially autonomous decision-making and interaction, without phoning home to a data center for every inference.

Why local AI matters more than a cute robot

The Apache 2.0 licensing is significant. Previous Gemma releases carried more restrictive terms. Full Apache 2.0 means commercial use, modification, and redistribution are all on the table without Google’s permission.

Google releasing Gemma 4 with multiple size variants optimized for edge devices, combined with a robot platform with a bill of materials under $400 running multimodal AI locally, is a proof of concept for deployments where cloud connectivity isn’t guaranteed or desirable.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.