▲ 465 today1.4k starsPythonNew · 6d old
Humanoid Motion Tracker Trains on a Single Differentiable World Model
What is supermimic?
SuperMimic learns to track a full-body humanoid’s pose by training a world model that can be differentiated end-to-end. It runs on MuJoCo-based mjlab and provides scripts to train and replay policies.
Who it is for
Robotics researchers and developers who need realistic whole-body motion estimation should look at it for fast prototyping.
Try it in 60 seconds
You need: CUDA-compatible GPU
curl -LsSf https://astral.sh/uv/install.sh | shmkdir supermimic_ws && cd supermimic_wsgit clone git@github.com:HybridRobotics/supermimic.gitgit clone git@github.com:HybridRobotics/mjlab_research.git
Apache-2.0 licenseUpdated 4 days agoPython23 open issues
New projects are not vetted. Skim the code, and run it in a fresh folder or a container before giving it your files or keys.
The story so far
Early Signal first spotted HybridRobotics/supermimic on November 12, 2026, when it had 1.3k GitHub stars. It has since grown to 1.4k. On its busiest day it gained 465 stars. The project is 6 days old and written mainly in Python. It is published under the Apache-2.0 license and has 114 forks. Early Signal files it under Robotics.
Questions about supermimic
- Is supermimic free and open source?
- Yes. HybridRobotics/supermimic is open source under the Apache-2.0 license, and the code is on GitHub at https://github.com/HybridRobotics/supermimic.
- How do I install supermimic?
- Use one of these, depending on your setup: curl -LsSf https://astral.sh/uv/install.sh | sh · mkdir supermimic_ws && cd supermimic_ws · git clone git@github.com:HybridRobotics/supermimic.git · git clone git@github.com:HybridRobotics/mjlab_research.git
- What do I need to run supermimic?
- CUDA-compatible GPU