CuboiLeo/SuperMimic-Reproduced
Reproduction of SuperMimic (arXiv:2610.28480): learned differentiable world model + BPTT policy for G1 humanoid motion tracking, on mjlab/MuJoCo Warp
Older snapshot: check the original source for current information.
From the publisher
README & documentation
Source previewRead the project’s overview, installation instructions and usage examples. The original README is the source of truth.
An independent, from-scratch reproduction of SuperMimic: Scaling Humanoid Whole-Body Control with One Differentiable World Model (arXiv:2610.28480, CoRL 2026) for the Unitree G1, built on mjlab / MuJoCo Warp. No official code is released; everything here (motion library, differentiable FK, world model, BPTT trainer, return-based reset sampling, data-parallel training, Appendix-F benchmark) is implemented from the paper. One policy, trained with the paper's observation layout on the full BONES-SEED set (129.9k clips) in 7.1 h on 4×RTX 5090. Solid robot: the policy in simulation (MuJoCo Warp, deterministic Appendix-F protocol). Blue ghost: the reference motion. Clips: high jump, getting up from lying face-down, 360° spin, one-leg hopping, a jog into a 270° turning jump and a drop to a crawl, inchworm. A longer cut with 18 motions (also dance turns, frog jumps, push-ups, kneeling, moonwalk, …) is in media/g1boneseedshowcase.mp4. Evaluation follows the paper's Appendix F: every clip is played once from its first…
Read the full README ↗ · Preview checked 2026-10-27T12:21:23.036Z
Inside the original README — Document outline
- SuperMimic-Reproduced
- Headline results
- What is implemented
- Setup
- Data
- Training
- Evaluation and media
- Repository layout
- Documentation
- Engineering notes
- Attribution and license
- 中文版
Headings are captured from the source. Links open the publisher’s document, not a locally hosted copy.
Links from the README
References supplied by the publisher, not independently verified endorsements. Check the destination before downloading files or entering credentials.
Documentation belongs to its respective authors. Reported project/model license: MIT. A listing is not a grant of reuse or training rights. Confirm the document’s own terms at the source.
What this repository does
Reproduction of SuperMimic (arXiv:2610.28480): learned differentiable world model + BPTT policy for G1 humanoid motion tracking, on mjlab/MuJoCo Warp
Repository facts
- Owner
- CuboiLeo
- Primary language
- Python
- Stars
- 295
- Forks
- 32
- Open issues + pull requests
- 0
- License
- MIT
- Archived
- No
- Default branch
- main
- Created
- 2026-10-21T23:36:13.000Z
- Last push
- 2026-10-26T21:21:54.000Z
Topics and intended use
No topics were included in the latest source metadata.
Review the README for scope, installation, examples and limitations. We do not run repository code or certify it.
Evaluate before installing
Review licensing and dependencies, inspect recent commits and unresolved issues, and test in an isolated environment before production use. Stars and forks alone cannot answer these questions.
Source and freshness
Source: GitHub. Metadata observed 2026-10-28T12:14:58.607Z. Daily imports are snapshots, not real-time monitoring.
Popularity and source listings do not establish security, suitability, licensing rights or benchmark performance.
Related repositories
EbookFoundation/free-programming-books
donnemartin/system-design-primer