Ooju provides a data infrastructure that captures high‑fidelity, multimodal 3D interaction data using portable XR devices, recording human hand motions, intent, and semantic context in real‑world environments. The automatically labeled, hardware‑agnostic datasets enable sample‑efficient training of dexterous manipulation policies that transfer between simulation and physical robots, supporting robotics and AI labs developing robust autonomous manipulation systems.
Funding
Funding not disclosed
Founders
Product
Problem
Robotic systems often rely on limited, lab‑generated datasets that lack the diversity and richness of real‑world interactions, leading to poor generalization when deployed in human environments.
Solution
Ooju creates a data infrastructure that captures high‑fidelity, multimodal interaction data using portable XR devices. By recording human hand motions, intent, and semantic context in both semi‑controlled and open‑world settings, the platform generates dense 3D datasets with automatic labeling. This data enables more sample‑efficient training of dexterous manipulation policies that can be transferred between simulation and physical robots. The approach is hardware‑agnostic, allowing the same dataset to support a variety of robot platforms without bespoke teleoperation setups. Ultimately, Ooju’s pipeline supplies the scale and variety needed to develop generalist agents capable of robust real‑world manipulation.
Target Audience
Primary customers are robotics research labs, AI labs, and companies developing autonomous manipulation systems that require large‑scale, high‑quality real‑world interaction data.
Features
- Portable XR capture system that records hand trajectories, depth, and intent in 3D space
- Automatic semantic labeling of actions and objects to produce ready‑to‑train datasets
- Multimodal data streams (vision, depth, pose, intent) aligned for simulation‑to‑real transfer
- Hardware‑agnostic pipeline that works with any robot platform for downstream learning
- Scalable collection process enabling diverse interactions across varied environments
- Dataset export formats compatible with common robotics simulation and training frameworks