GitHub radar
Microduck: Real AI Robot Policies in Browser
Microduck with RL-trained control policies from Pollen Robotics, hosted online as browser-based sandbox without installation! WebAssembly physics engine, onnxruntime-web model, multiplayer, no-install.
The French robot company Pollen Robotics has published a web simulator of their Microduck humanoid robot using their pollen-robotics namespace on Hugging Face. This uses real reinforcement learning policies learned on physical Microduck robots, not pre-recorded videos. The control policies are implemented with a neural network that is run using onnxruntime-web in the browser, there is no need to upload your data to some server, it just runs locally in your browser tab. There are two different kinds of locomotion (wheels & legs) and 6 predefined behaviours (running, sitting, somersault, ball kicking, roller skating, squatting). If you open the web app in your browser you’ll see multiple visitors as transparent ghosts running around in the same scene at the same time.
Why a vibe-coder should care
This is a great demonstration of how to run heavy AI algorithms on the web using WebAssembly, without having to use Python.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Open this simulator in a browser: https://huggingface.co/spaces/pollen-robotics/microduck-simulator — try a few behaviors and explain how the AI motion control is implemented here.
Nothing to install — runs in the browser.
Open on Hugging Face▌ More finds