Gemma Playground: Robot Duck
Key Takeaways
Builds a robot duck using Gemma 4 E2B on LiteRT and Jetson Orin Nano with multimodal inputs
Original Description
Xavier Plantaz, Partner Solutions Engineer at Google, brings two Open Duck Mini v2 robots, built by Antoine Pirrone, on-device with Gemma 4. One runs Gemma 4 E2B on LiteRT on a Raspberry Pi 5. The other runs Gemma 4 E2B on a Jetson Orin Nano. Microphones, cameras, speakers, and Gemma's multimodal inputs let each duck listen, look, and talk back in real time.
What's covered: Running Gemma 4 E2B on Raspberry Pi 5 and Jetson Orin Nano, the on-device voice stack (Parakeet for speech-to-text, Gemma 4 for inference, Kokoro for text-to-speech), LED and antenna expressiveness, attention mode and live conversation, and where the project goes next as the ducks start to see and talk to each other.
Explore the Open Duck Mini v2 project on GitHub and try running Gemma 4 on your own hardware.
What are you building on-device with Gemma? Drop it in the comments.
Subscribe to Google for Developers → https://goo.gle/developers
Speaker: Xavier Plantaz
Products Mentioned: Google AI, Gemini, Gemma
Watch on YouTube ↗
(saves to browser)
Sign in to unlock AI tutor explanation · ⚡30
More on: Agent Foundations
View skill →Related Reads
📰
📰
📰
📰
OpenAI makes ChatGPT Health available to all U.S. users
TechCrunch AI
I built an AI-first bank statement converter. The one rule that stopped it lying to me.
Dev.to AI
Building trade assistant: How Jefferies optimized front office trading operations with AI
AWS Machine Learning
One Database for the Whole LangChain Ecosystem: Memory, Persistence, and Deep Agents on Oracle AI Database
Dev.to · Anya Summers
🎓
Tutor Explanation
DeepCamp AI