High-Quality Spatial & Physical Intelligence Data for Model Builders
Real-world data created by natural human activity. Select a domain below to learn more.
Spatial & Physical Intelligence
High-quality, real-world training data to help bring AI into the physical world. Whether you’re building in robotics, world models, or something else, access and curate the right combination of egocentric video, motion capture, video game data, and 3D assets from our library.
Learn more about our approach to Spatial & Physical Intelligence
First-person visual narratives for tasks and navigation
LiDAR, Depth, Hand-tracking
Broad & Diverse Motion Capture Content
50k+ Unique Performers
Human input mapped to real-time environment response
Human-Object-Environment Interaction
Semantic-rich Geometric Scans of Objects and Spatial Layouts
Millions of Digital Twins
Human–Object Interactions
Household and industrial motions like opening doors, handling tools, cooking, and manipulating everyday objects for robotics and embodied agents.
Locomotion and Navigation
Walking, running, turning, stairs, and varied terrain that capture how people actually move through real environments.
Gestures and Body Language
Conversational hand movements, posture shifts, and expressive full-body cues used in social interaction.
Real-World Object Archives
Millions of authentic 3D scans across reflecting the diversity of objects encountered in daily life around the world.
Sports & Athletic Actions
High-intensity, dynamic movements from field, court, and individual sports.
Combat & Stunt Motion
Martial arts, falls, impacts, and complex stunt choreography for high-fidelity physical interactions.
Interactive Game Data
Behavioral and movement sequences from high-fidelity digital worlds, from player inputs to resulting pixel-level transformations.
Environmental Scenes
3D scans and geometric data of household, industrial, and public spaces, providing the environmental scale required for spatial reasoning.
Studio-Grade Reference Motion
High-quality, well-instrumented captures suitable for pre-training, benchmarking, and fine-tuning motion models.