Five-Finger Data Collection Glove

Sensing the Myriad Realms, Awakening Embodied AI

Standardized and unified structure, consistent data baselines, efficiently capturing hand motions of diverse operators and amassing high-consistency, reusable high-quality data.

Multidimensional array tactile sensor

Full coverage across fingertips and palm

Number of tactile signals: 1,743

3 Dimensional Force + Distributed Force

0.1 N Tactile Force Sensitivity

1% FS Repeatability

Wide-FOV Wrist Camera

Complete Capture of Hand Manipulation and Hand-Object Interaction Details

165° FOV

1080p Resolution

Global Shutter

Miniature High-Precision Magnetic Encoder

In-House Developed Miniature High-Precision Angular Encoder

Magnetic Interference Resistant, Zero-Drift Pose

Equipped with 17 Angular Encoders

6.5 mm Diameter

±0.15° Accuracy

Anti-Magnetic Interference (< 0.2°)

High-Quality Data Collection Paradigm

High-Consistency Data Standards and Specifications

Capture real human hand experience under unified standards, building a highly consistent and precise data foundation for robot training.

Omni-Modal, Multi-Source Sensing Fusion Capture

Integrated with tactile sensing, vision, joint-state, and hand pose signals to comprehensively capture human hand behaviors, tailored for embodied AI model training and deployment.

Scalable and Efficient Deployment

Accelerating Scalable Data Collection

Co-optimized process and hardware design , enables long-term, large-scale, and routine data collection.

High Usability & Standardized Interfaces

Simple interaction and universal interfaces enabling plug-and-play use across diverse scenarios.

Lightweight, Human-Conformal Design

Lightweight & Easy-to-Don

Lightweight ergonomic design for authentic, efficient and comfortable hand data collection.

Bio-Inspired Conformal Human-Driven Dexterous Hand

Ergonomically fitted to the human hand, the glove enables precise operation and smooth control.

Ready-to-Use Data on Acquisition

Hardware-Processed Raw Data with Direct Output

Sensors capture raw signals, while built-in algorithms process them into ready-to-use formats—delivering training-ready data out of the box and shortening development cycles.