Human motion data
for Physical AI.
A teaser gallery of multimodal human behavior datasets captured with e-skin MEVA — motion, video, pressure and context for Physical AI.
Why MEVA?
MEVA captures accurate full-body motion and packages it with multimodal context for Physical AI-ready humanoid learning datasets.
Physical AI Ready!
Motion, contact, vision, pressure, and body metadata are organized as reusable human behavior data for humanoid learning pipelines.
| Item | Monocular AI | Multi-View Markerless | Simplified IMU | MEVA |
|---|---|---|---|---|
| Capture method | Camera-based | Camera-based | IMU-based | IMU-based |
| Sensors / cameras | 1 camera | Multiple cameras | Approx. 6–12 IMUs | 18 IMUs |
| Full-body mocap accuracy | Bad Insufficient positional accuracy from monocular depth estimation | Good Accurate 3D reconstruction using calibrated multi-view cameras | Bad Not enough IMUs to cover the minimum 15 full-body segments | Good Covers the required full-body segment structure |
| Ability to capture unseen or non-standard motion | Limited | Good | Limited | Good |
| Dedicated studio requirement | No | Yes | No | No |
| Real-world / field usability | Medium | Limited | Good | Good |
| Setup complexity | Very low | High | Low | Very low |
| Approx. system cost | Free / software-based | US$60k–200k+ | Approx. US$2k | Approx. US$20k |
| Effectiveness for Physical AI | Limited | Good but studio-bound | Limited | Very good |
For full-body humanoid learning, 15 IMUs is the practical minimum for direct segment coverage. MEVA uses 18 IMUs and adds video, ground-contact probability, pressure, and metadata as Physical AI-ready datasets.
MEVA Omni Capsules
Each Omni Capsule packages motion, vision, contact and context as reusable human behavior data for Physical AI.
Loading capsules...
Need a motion that is not listed here? We can capture new human behavior data with MEVA.
Multimodal Dataset
Beyond motion, vision, contact and context. Multimodal behavior data gives Physical AI the ability to perceive.
Motion
Skeleton, BVH, CSV and future SMPL-X style formats for movement learning.
Vision
Third-person and future EgoVision previews to preserve visual context.
Pressure
Foot, seat or environment pressure maps connected to human movement.
Robot-ready
Clean metadata, labels and downloadable packages for Physical AI pipelines.