Microduck RL
What it is
A reinforcement learning training environment for the Microduck bipedal robot, built on the mjlab (MuJoCo Warp) framework. It provides a full sim2real pipeline including high-fidelity actuator modeling, domain randomization, and PPO training for various locomotion and manipulation tasks.
The problem it solves
Standard RL environments often fail to capture the specific physical dynamics (like backlash and friction) of small-scale servos, leading to policies that fail when deployed on real hardware.
What you could use it for
- Training walking and velocity-tracking policies for small bipedal robots
- Simulating and training fall recovery and stand-up behaviors
- Exporting trained policies to ONNX for real-robot deployment
- Researching sim2real transfer techniques using detailed actuator physics
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.