Physicl provides an automated pipeline that turns raw visual inputs—images, video, scans, or 3D files—into simulation‑ready 3D models with physics attributes such as mass, friction, and collision data. The platform cleans geometry, derives material properties, generates parametric scene variations, and exports assets in USD, URDF and other formats compatible with simulators like Isaac Sim, MuJoCo, and Habitat, enabling robotics and AI teams to scale their training datasets.
Funding
Funding not disclosed
Founders
Product
Problem
Physical AI models require large volumes of simulation-ready 3D assets that include accurate physics properties such as mass, friction, and collision data. Existing asset libraries often lack these annotations or need extensive manual cleanup, creating bottlenecks for robotics and foundation model training.
Solution
Physicl offers an end‑to‑end data infrastructure that converts any raw visual input—images, video, scans, or 3D files—into structured, physics‑tagged 3D models ready for simulation. The platform automatically extracts geometry, material attributes, and derives physics parameters, then applies parametric generation to produce infinite scene permutations with controlled variations in lighting, layout, and object scale. Every asset undergoes multi‑level human validation, achieving a 98% simulation‑ready rate before release. Export formats include USD, URDF, and render outputs compatible with major simulators such as Isaac Sim, MuJoCo, and Habitat, enabling seamless integration into training pipelines at scale.
Target Audience
Primary customers are robotics research teams, AI labs developing foundation models, and simulation engineers who need large, physics‑accurate 3D datasets for training and testing.
Features
- Automatic ingestion of any image, video, scan, or 3D file without pre‑processing
- Normalization pipeline that cleans geometry, resolves materials, and derives friction, mass, and collision properties
- Parametric scene generation for unlimited variations in lighting, layout, occlusion, and object scale
- Human‑validated quality control with over 10,000 3D specialists achieving a 98% QC pass rate
- Export of assets in USD, URDF, and high‑resolution render formats compatible with Isaac Sim, MuJoCo, Habitat, and other simulators
- Built‑in metadata including semantics, graspability, and navigation meshes for both static and dynamic objects