cheryl-tootty/mm-motion-v1

MM-Motions

MM-Motions is a reproducible MyoFullBody motion collection assembled for MuscleMimic. Release status: draft.

The frozen splits contain:

• train: 31974 motions (10346 PHC and 21628 BONES)
• validation: 875 motions (283 PHC and 592 BONES)
• test: 2630 motions (850 PHC and 1780 BONES)
• total duration: 95.17 hours across 35479 verified motions

The builder excluded 0 source motions that failed the frozen GMR quality gates. Their identities and metrics are recorded in quality_rejections.jsonl.

PHC retains its frozen quality-filtered source splits. BONES uses the same train, validation and test proportions, assigned with deterministic seed 42. BONES original and mirrored variants remain in the same release split.

Every manifest record identifies its source motion, original source split, retargeted payload SHA-256, payload size and GMR cache path. Verified payloads also record shape, frame count and sampling frequency.

Payload provenance in this release:

• PHC retargeting code revision: unknown
• BONES retargeting code revision: unknown
• BONES retargeting condition: shape_aligned
• BONES fitted artifact SHA-256: 4b7911548a0823caeaa9c579da2703616c9955c7635cae753caaf27a78a0aca0

License and access

MM-Motions combines outputs derived from AMASS and BONES-SEED. Their source licenses remain controlling. This repository must not host payload files until the release metadata records written redistribution permission from both rights holders. Motion-level manifests also expose source identifiers and need license review before public distribution. A public landing page may describe the collection, access workflow and citation without including those records. Users must obtain source access from each provider and comply with its terms.

• AMASS license: https://amass.is.tue.mpg.de/license.html
• BONES-SEED license: https://bones.studio/info/seed-license

Motion Data by Bones Studio: https://bones.studio/

Citation

@article{Li2026MuscleMimic,
  title={Towards Embodied AI with MuscleMimic: Unlocking full-body musculoskeletal motor learning at scale},
  author={Li, Chengkun and Wang, Cheryl and Ziliotto, Bianca and Simos, Merkourios and Kovecses, Jozsef and Durandau, Guillaume and Mathis, Alexander},
  journal={arXiv preprint arXiv:2603.25544},
  year={2026}
}

Users must also cite AMASS, its applicable constituent datasets and BONES-SEED as required by their licenses.