chengkunli's picture
Update README.md
7709880 verified
metadata
license: other
task_categories:
  - robotics
tags:
  - motion-retargeting
  - musclemimic
  - mujoco
  - amass
pretty_name: MuscleMimic GMR-Fit Retargeted Motions
extra_gated_heading: Agreement required before accessing this dataset
extra_gated_description: >-
  This dataset is derived from AMASS and is provided only for permitted
  non-commercial research uses.
extra_gated_button_content: I agree and request access
extra_gated_prompt: >
  By requesting access to this dataset, you acknowledge and agree that:


  1. This dataset is derived from AMASS data and is subject to restrictions
  consistent with the AMASS license.

  2. You will use this dataset only for non-commercial scientific research,
  non-commercial education, or non-commercial artistic projects.

  3. You will not copy, share, distribute, sublicense, transfer, or otherwise
  make this dataset available to any third party.

  4. You will not use this dataset, or any models/methods trained on it, for
  commercial purposes of any kind.

  5. You will not redistribute this dataset or derivative files generated from
  it unless you have obtained all necessary permissions from the relevant rights
  holders.

  6. You understand that access may be revoked if these conditions are violated.

  7. You agree to appropriately cite AMASS and our work in any resulting
  publications.
extra_gated_fields:
  Full name: text
  Email: text
  Affiliation: text
  Intended use:
    type: select
    options:
      - Non-commercial scientific research
      - Non-commercial education
      - Non-commercial artistic project
      - Other
  I confirm that I will use this dataset only for non-commercial purposes: checkbox
  I confirm that I will not redistribute this dataset or derivative files to any third party: checkbox
  I confirm that I will not use this dataset or models trained on it for commercial purposes: checkbox
  I confirm that I have read and agree to comply with the original AMASS license terms: checkbox

MuscleMimic GMR Retargeted Motions

Pre-retargeted motion capture data for the MyoFullBody musculoskeletal model, generated using General Motion Retargeting (GMR).

Dataset Groups

Group Motions Description
KIT_KINESIS_TRAINING_MOTIONS 972 KIT locomotion training set
KIT_KINESIS_TESTING_MOTIONS 108 KIT locomotion test set
AMASS_TRANSITION_MOTIONS 110 AMASS action transitions

Usage

# Auto-download (integrated into training pipeline)
# Just set retargeting_method: gmr in your config - caches download automatically.

# Manual download
uv run musclemimic-download-gmr-caches --dataset-group KIT_KINESIS_TRAINING_MOTIONS

# Python API
from musclemimic.utils import download_gmr_dataset_group
download_gmr_dataset_group("KIT_KINESIS_TRAINING_MOTIONS")

File Format

Each .npz file contains retargeted motion data with arrays: qpos, qvel, xpos, xquat, cvel, subtree_com, site_xpos, site_xmat, plus metadata (joint_names, body_names, site_names, frequency, split_points).

Source

Derived from the AMASS dataset. Please comply with the AMASS license terms when using this data.

Paper: https://huggingface.co/papers/2603.25544

bibTeX

@article{li2026towards,
  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}
}