microduck
SHA-256Open-source brain of a 25 cm biped duck robot that walks, rolls, and recovers from falls using reinforcement-learning policies, with a production-grade Rust daemon stack and full MuJoCo simulator.
Smart Download
Download Download Version
vdaemon-v0.17.8 · 10.3 MB
An open-source 25 cm biped duck robot that walks, rolls, picks things up, and stands itself back up using reinforcement-learning policies.
Core Features
- RL-driven locomotion: walking, roller mode with wheels, object pickup, and autonomous fall recovery
- Production-grade Rust daemon stack running a 50 Hz control loop over fifteen servos
- Signed, health-gated OTA updates with automatic rollback so a bad release never bricks the robot
- Full MuJoCo simulator (duck-sim) running the real daemons—develop and test without hardware
- Multiple control paths: gamepad, Bluetooth from phone or laptop, and WebRTC camera streaming
What It Can't Do
- •This is not a standalone app: the full experience requires the physical robot (sold separately) or a self-built RK3566 board. Walking policies are trained in the separate microduck_rl repo—this repository only runs them. Gamepads occasionally fail to bond and need manual re-pairing per the docs. The stack is deliberately ROS-free and Rust-centric, so expect a learning curve if you come from traditional robotics frameworks.
Use Cases
- Learning the complete sim2real reinforcement-learning pipeline on affordable real hardware
- Extending the robot through its unified JSON-RPC API shared by every client
- Teaching bipedal locomotion and embedded Rust systems in labs and classrooms
- Hacking on a desk-friendly robot pet that walks, kicks a ball, and quacks
Detailed Introduction
Microduck is the open-source brain of a tiny biped duck robot—about 25 cm tall and 800 g—built by Pollen Robotics. A handful of Rust daemons on a Rockchip RK3566 run a 50 Hz control loop driving fifteen servos with neural policies trained via PPO in MuJoCo and deployed as ONNX. The duck walks with a gamepad, switches to roller mode with wheels attached, picks objects off the floor, and stands itself back up after a fall. Unlike most hobby robot kits, Microduck ships a production-grade software stack: signed and health-gated OTA updates with automatic rollback, a unified JSON-RPC API over Unix sockets shared by every client, WebRTC camera streaming, and a MuJoCo simulator (duck-sim) that runs the real daemons so you can develop without hardware. The walking policies live in the companion microduck_rl repo, which documents the complete sim2real recipe. Compared with platforms like Stanford Doggo or Unitree's entry-level robots, Microduck is far smaller, cheaper, and fully open down to the update machinery—an unusually accessible way to learn reinforcement-learning robotics on real hardware.
Tags
Getting Started
Install the software
Install the appropriate package for your distro (dpkg / rpm / AppImage)
Get a Microduck from the official site, or flash a Rockchip RK3566 dev board following docs/robot/install-dev.md
Power on the robot and pair a gamepad once, then check the cheat sheet for the full button mapping
Drive it with the gamepad, or connect from your phone or laptop over Bluetooth using duckctl
- Get a Microduck from the official site, or flash a Rockchip RK3566 dev board following docs/robot/install-dev.md
- Power on the robot and pair a gamepad once, then check the cheat sheet for the full button mapping
- Drive it with the gamepad, or connect from your phone or laptop over Bluetooth using duckctl
SHA-256 checksum verified
Checksum extracted from GitHub official Release page
SHA256 Checksum
b5adc045fd16610524e976e80f5a0c4130a193aeef88d08267c323891718de26This checksum is extracted from the GitHub Release page. Verify file integrity after download.
All SHA-256 checksums on this platform are extracted from the project's official GitHub Release page, without any modification. You can independently verify them on the GitHub Releases page.
Open Source Transparency
View GitHub SourceUninstall Info
Microduck is robot firmware, not a desktop app. Developers can simply delete the cloned repository. On the robot itself, use the robotctl rollback command to revert to the previous release, or reflash the official image to reset the system completely.
No Extra Dependencies
Ready to use after download. No additional runtime required.
Similar Projects
ollama
Ollama lets you download, run, and manage large language models locally. One command, multiple platforms, endless possibilities.
ComfyUI
ComfyUI is a modular AI engine with a node-based graph interface for generating images, video, 3D models, and audio. Unlike traditional webui tools, it gives pixel-level control over every model and parameter, making it the preferred choice for professionals.
llama.cpp
High-performance LLM inference engine in C/C++ with minimal dependencies, supporting quantized models (1.5–8 bit) and diverse hardware (Apple Silicon, CUDA, Vulkan, etc.).