License 250+ hours of full-body human motion captured at 120 fps on a calibrated 120-camera Vicon system—the same production stage used for AAA games and Hollywood work. The data is cleaned, segmented, labeled, and prepared for model training and evaluation.
If a model needs to learn how people move, the source matters. This dataset begins with optical markers tracked throughout a calibrated capture volume—not poses reconstructed from ordinary video. The result is ground-truth 3D skeletal motion with the temporal continuity needed for motion generation, motion understanding, and evaluation.
The 120-camera Vicon system captures at 120 fps with submillimeter marker-tracking precision. Dense camera coverage is designed to reduce occlusion, while the capture rate preserves fast actions, subtle transitions, and full-body coordination over time.
The motion was recorded on the same capture stage used to produce performance data for AAA games and Hollywood projects. The facility combines calibrated optical tracking, finger capture, synchronized reference systems, and a professional post-processing workflow.
That production pedigree matters because capture quality is more than a camera count. It depends on a controlled volume, reliable calibration, performer direction, reference media, careful solving, and human review. The same disciplines used to create motion for high-end entertainment production are applied to data intended for AI training.
Raw motion capture still requires substantial work before it is useful to a model team. Uthana prepares the licensable dataset as structured training material, with human review applied to the motion and its annotations.
Human-reviewed motion cleanup and quality assurance
Clips divided at action boundaries
A mixture of human- and machine-written labels or text descriptions, all human-reviewed
Standardized skeleton and coordinate conventions
Root trajectories, joint trajectories, and foot-contact annotations
Train, validation, and test splits
This preparation makes it easier to inspect coverage, select relevant motion, build repeatable data loaders, and evaluate models against a documented source.
The dataset spans both functional movement and expressive performance. It includes motion across different genders and ages, helping teams train and evaluate against more than one body or movement style.
The exact catalog and category coverage can be reviewed during dataset evaluation. Object-interaction motion is included. Hand-object contact labels can be added to the standard annotation set as an add-on.
Finger motion is included across the dataset, preserving more of the performance than a body-only skeleton. Uthana documents the source representation, skeleton hierarchy, coordinate conventions, units, frame rate, segmentation, and annotation fields supplied with a delivery.
The target skeleton and delivery format can be customized to the engagement. Potential targets include production character skeletons and humanoid embodiments such as Unitree G1. The delivery specification is agreed before licensing so the representation fits the intended training pipeline.
Review data specificationsThe dataset has been used in Uthana's own motion-model development and by an external team for model training. This provides downstream evidence that the data is organized for model work, not only repackaged animation content.
Review a representative sample before deciding whether the dataset fits your training or evaluation task. Uthana can provide the current motion catalog, a sample of the data and annotations, and the proposed delivery specification through a gated request.
Commercial training and evaluation rights are available by agreement. Dataset scope, target representation, delivery format, and permitted use are defined as part of that engagement.
The source motion is captured with optical markers tracked inside a calibrated Vicon volume. The delivered skeletal motion is solved from those marker observations rather than inferred solely from ordinary video pixels. Human-reviewed cleanup and QA are applied before delivery.
The dataset includes cleaned full-body skeletal motion, finger data, action-boundary segmentation, action labels or text descriptions, standardized skeleton and coordinate conventions, root and joint trajectories, foot-contact annotations, metadata, and train, validation, and test splits. The exact field specification is documented for the licensed delivery.
Yes. Finger motion is included across the dataset, not limited to a separate subset.
Action labels and text descriptions are produced through a mixture of human and machine writing. All are human-reviewed before they are included in the prepared dataset.
Coverage includes locomotion and transitions, everyday actions and gestures, dance and expressive performance, athletics and sports, combat and stunts, and object manipulation. Request the current catalog to inspect the exact actions and available duration in each category.
The target skeleton and delivery format can be customized to the engagement. Uthana confirms the source and target representations, coordinate conventions, units, frame rate, file structure, and requested metadata before delivery.
Yes. Request a representative sample through sales@uthana.com. Uthana can also provide the current catalog and a proposed delivery specification for evaluation.
Commercial training and evaluation rights are available by agreement. The permitted uses and delivery terms are defined for each engagement.
Start generating motion today, or work with us on a custom dataset, embodiment system, or simulation program.
