High-Fidelity Human Motion Capture Dataset for Humanoid Robotics
Human-Robot Interaction Data
Tags and Keywords
Free
About
Apple Arts Studios Motion Capture Dataset
Dataset Overview
The Apple Arts Studios Motion Capture Dataset is a professionally captured, full-body human-motion dataset developed for applications including:
- Artificial Intelligence
- Humanoid Robotics
- Physical AI
- Motion Generation
- Simulation
- Animation
- Action Recognition
- Human-Motion Research
This dataset release contains 5 hours of originally captured human motion data.
Left-right mirrored versions of the original motions are also included, increasing the total distributed motion duration to 10 hours.
Original captured motion: 5 hours
Mirrored motion augmentation: 5 hours
Total distributed motion duration: 10 hours
The mirrored motions are derived from the original recordings and should be treated as data augmentation, not as independently captured performances.
The dataset includes full-body human movement captured across multiple environments and interaction conditions, including:
- Flat rubbered floors
- Ramps
- Stairs
- Object-interaction environments
The motion library includes multiple:
- Movement speeds
- Directions
- Body positions
- Transitions
- Movement styles
- Performer variations
Each motion is provided using structured representations suitable for AI, robotics, simulation, and motion-processing pipelines.
Included Representations
- SOMA Uniform BVH at 30 FPS
- Unitree G1 CSV
- JSON metadata
- Original and mirrored motion variants
- Structured environment, category, action, and actor hierarchy
Dataset Origin
This dataset was conceived, performed, captured, processed, and curated entirely in-house by Apple Arts Studios at our own professional motion-capture facility.
The dataset was developed for internal research, dataset development, humanoid robotics, Physical AI, motion-generation, simulation, and related human-motion applications.
Dataset Tiers
Apple Arts Studios maintains motion data in two primary dataset tiers.
Tier 1 — Segmented Motion Clips
Tier 1 consists of individually trimmed, cleaned, and training-ready motion clips.
Each file contains one clearly defined action, separated by performer and variation, with T-poses, idle periods, and unrelated motion removed.
Tier 1 Specifications
- Typical clip duration: 10–20 seconds
- SOMA Uniform BVH: 30 FPS
- Unitree G1 representation: CSV
- Metadata: Structured CSV and JSONL
- Dataset release included in this listing:
- Original: 5 hours
- Mirrored: 5 hours
- Total: 10 hours
Full Tier 1 Library
- Original: 67 hours 04 minutes
- Mirrored: 67 hours 04 minutes
- Total: 134 hours 08 minutes
Tier 2 — Full-Length Capture Sequences
Tier 2 consists of continuous motion-capture recordings preserving the complete capture timeline.
These sequences may include:
- T-poses
- Idle actions
- Movement transitions
- Performer resets
- Repositioning
- Settle time
- Complete motion variations
Full Tier 2 Library
- Original: 195 hours 48 minutes
- Mirrored: 195 hours 48 minutes
- Total: 391 hours 36 minutes
Synchronized witness-camera footage and OptiTrack Motive reference recordings are also available for the original capture sessions.
Important: Tier 1 clips are segmented from Tier 2 recordings and are not additional capture hours. Mirrored files are left-right augmented versions of the original motion and are not separately captured performances. Mirroring applies only to motion data and not to the reference recordings.
Dataset Information
| Property | Description |
|---|---|
| Dataset Provider | Apple Arts Studios |
| Dataset Type | Full-body + fingers human motion capture |
| Original Capture Duration | 5 hours |
| Mirrored Duration | 5 hours |
| Total Distributed Duration | 10 hours |
| Total Files | 2,584 original and mirrored files combined |
| Retargeted BVH Representation | SOMA Uniform BVH |
| SOMA Uniform Frame Rate | 30 FPS |
| Humanoid Robot Representation | Unitree G1 |
| Unitree G1 Format | CSV |
| Metadata Format | JSON |
| Motion Variants | Original and mirrored |
| Dataset Organization | Environment, category, action, and actor |
| Primary Applications | AI, humanoid robotics, Physical AI, animation, simulation, and motion research |
Motion Representations
1. SOMA Uniform BVH
The SOMA Uniform BVH files contain the captured human motions converted to a standardized SOMA Uniform skeletal representation.
These files are provided at 30 FPS and use a consistent:
- Skeleton hierarchy
- Joint naming convention
- Bone structure
- Body proportion
- Reference representation
Using a standardized skeletal representation allows motions from different performers to be compared and processed using a consistent structure.
The SOMA Uniform representation may support applications such as:
- Motion-generation research
- Motion retrieval
- Skeletal representation learning
- Cross-actor motion comparison
- Standardized animation processing
- Human-motion machine learning
- Motion classification
- Motion prediction
- Motion embedding research
2. Unitree G1 CSV
The Unitree G1 CSV files provide human motion data mapped to a Unitree G1 humanoid robot representation.
These files are intended for robotics research and development workflows and may support:
- Human-to-robot motion retargeting
- Humanoid imitation learning
- Motion-policy development
- Robotics simulation
- Simulation demonstrations
- Whole-body control research
- Physical AI
- Reinforcement-learning research
- Robot-motion visualization
Safety and Compatibility Notice
The Unitree G1 CSV files should not be treated as immediately deployable robot-control commands.
Users are responsible for validating motion compatibility with their specific robot, simulator, controller, joint constraints, collision model, balance system, dynamics model, safety system, and deployment environment.
3. Structured Metadata
Each motion is accompanied by structured metadata.
Metadata may include:
- Motion identifier
- Motion name
- Action name
- Category
- Subcategory
- Actor identifier
- Take number
- Version number
- Motion representation
- Frame rate
- Frame count
- Duration
- Original or mirrored status
- Movement direction
- Movement speed
- Object or prop information
- Relative file path
- Capture-system information
- Skeleton information
The metadata allows corresponding SOMA Uniform BVH and Unitree G1 CSV representations of the same motion to be identified and paired.
Frame-Rate Information
The dataset contains representation-specific frame rates.
| Representation | Frame Rate |
|---|---|
| SOMA Uniform BVH | 30 FPS |
| Unitree G1 CSV | Recorded in the corresponding file metadata |
Users should not assume that all distributed representations use the same frame rate.
When comparing, processing, or synchronizing SOMA Uniform BVH and Unitree G1 CSV files, users should account for:
- Resampling
- Frame indexing
- Frame count
- Motion duration
- Timestamp alignment
- Start-frame alignment
- Representation-specific processing
Filenames remain consistent across representations so corresponding files can be identified and paired programmatically.
Original and Mirrored Motion Data
The dataset contains both original and left-right mirrored motion files.
Original Motions
Original motions are human performances captured and processed through the Apple Arts Studios professional motion-capture pipeline.
Mirrored Motions
Mirrored motions are derived from the corresponding original motion by applying a left-right motion transformation.
Mirrored files are identified in the metadata using their original or mirrored status.
Where applicable, mirrored motions remain connected to their original source motion through the corresponding source-motion identifier.
Mirrored motions are included as augmented data and are not counted as separately captured performances.
Dataset Organization
The dataset uses a structured hierarchy based on:
- Motion representation
- Motion environment
- Original or mirrored status
- Surface or interaction type
- Action category
- Action name
- Actor identifier
Example structure:
AAS_Motion_Dataset/
│
├── Soma_Bvh/
│ └── Locomotions/
│ ├── Original/
│ └── Mirrored/
│
├── Unitree_G1_Csv/
│ └── Locomotions/
│ ├── Original/
│ └── Mirrored/
Listing Stats
VIEWS
24
DELIVERY
INSTANT DOWNLOAD
LISTED
14/08/2026
UPDATED
14/08/2026
REGION
GLOBAL
TRUST
5 / 5
Loading...
Free
Download Dataset in ZIP Format
Recommended Datasets
Loading recommendations...