MOVAS-MoCap — Human Kinematic Dataset for Embodied Learning
Optical and inertial motion capture of full-body and hand kinematics during real manipulation tasks, aligned to the egocentric and multi-view corpora.
Overview
Human kinematics matter for humanoids because the kinematic chains are close enough that motion transfers with modest retargeting loss. MOVAS-MoCap captures that motion during genuine work rather than in a mocap studio.
A subset includes force-plate ground reaction data, which is the piece most commonly missing when teams try to learn loco-manipulation from video alone.
Data Structure
Directory layout as delivered. Every clip is self-contained: no cross-referencing required to train on a single sample.
movas-mocap/
├── takes/{take_id}/
│ ├── skeleton.bvh
│ ├── smplx.npz
│ ├── hands_mano.npz
│ ├── force_plate.npz # subset only
│ └── ref_video.mp4
└── manifest.parquetAnnotation Schema
Take
Task, subject anthropometrics, rig configuration.
Segment
Phase-segmented motion primitives.
Contact
Hand-object and foot-ground contact events.
Quality Gates
Published thresholds, enforced at ingest. Records that fail are rejected or flagged in the manifest — never silently included.
| Gate | Threshold | Method |
|---|---|---|
| Marker residual | < 0.8 mm | Optical system RMS across capture volume. |
| Gap-fill rate | < 2% frames | Interpolated frames flagged in the manifest. |
| Video sync | < 1 frame | Genlocked to reference cameras. |
Formats & Access
Export formats
Modalities included
Cloud delivery
Direct to your S3, GCS or OSS bucket. Manifest-driven incremental sync.
Air-gapped
Physical media transfer for regulated or offline environments.
Via Movas-OS
Stream and re-annotate in place through the Movas-OS platform.
Use Cases
Humanoid whole-body control
Reference trajectories for loco-manipulation policies.
Retargeting ground truth
Validate video-derived pose against optical ground truth.
Biomechanics-aware planning
Force-plate subset supports dynamics-consistent motion.