Microduck
Open-source mini bipedal robot for training your own RL policies
About Microduck
Microduck is a 25 cm-tall open-source bipedal robot that works out of the box. It supports training and retraining behavior policies on your own machine via reinforcement learning, learning motions in physical simulation, and completing training locally or with Hugging Face Jobs. The full software stack is Apache-2.0 licensed, the robot carries a camera, LiDAR, and two IMUs, with an onboard policy loop at 50 Hz, and it ships with a battery, a USB-C cable, and a game controller.
Best for
Best use cases
Retrain robot behaviors on your own machine
Learn policies in physical simulation
Exchange robot policies
Unbox and try the robot right away
Features
RL retraining
Every behavior is a policy you can retrain on your own machine.
Physical simulation training
Behaviors can be trained in MuJoCo physics simulation or with Hugging Face Jobs.
Open-source software stack
The full software stack is Apache-2.0 licensed.
Multi-sensor configuration
Equipped with a camera, LiDAR, and two IMUs.
Pros
- Works out of the box
- Supports retraining behavior policies locally
- Supports physical simulation training
- Full software stack is Apache-2.0 licensed
Pricing Plans
Microduck
- Robot, battery, USB-C cable, and game controller
* Pricing information is for reference only. Please refer to the latest announcement on Microduck's official website.