CLARU
flex-piapache-2.0

RoboTwin 2.0 3D — T5 Text Embedding Cache

Precomputed UMT5-XXL text embeddings for 1,039,891 unique task prompts from the RoboTwin 2.0 3D dataset, designed for use with the Wan2.2-TI2V-5B model.

Downloads0
Episodes1039891

Why This Matters for Physical AI

This embedding cache enables efficient training of vision-language models for robotic manipulation by eliminating expensive recomputation of text encodings, accelerating the development of instruction-following robotic systems.

Technical Profile

Modalities
language
Episodes
1039891
Data Format
safetensors
Annotation Types
language_instructions
License
apache-2.0
Part of the RoboTwin 2.0 3D family

Access

View on HuggingFace

Need custom language data?

Claru builds purpose-built datasets for any environment applications with dense human annotations and quality assurance.

Request a Sample Pack

Related Datasets

Robo-MLT: Medical Laboratory Manipulation Datasets

Real-robot teleoperated demonstrations for two long-horizon medical laboratory workflows (blood gas analysis and tube sorting) collected on an Agilex dual-arm Static-Aloha platform with vision-language-action supervision.

rgbproprioceptionlanguage
null downloadsJul 2026apache-2.0gated

HiFi-UMI-2K: High-Fidelity Robot-Free Manipulation Data

A large-scale dataset of 2,000 hours of high-fidelity bimanual human manipulation demonstrations collected robot-free using the HiFi-UMI capture system, designed for action-grounded supervision for manipulation policy pre-training and deployment without requiring a robot during data collection.

rgbimuproprioceptionlanguage
133K downloadsJul 2026cc-by-4.0

Xperience-10M

A large-scale egocentric multimodal dataset of human experience containing 10 million interactions and 10,000 hours of synchronized first-person recordings with six video streams, audio, stereo depth, camera pose, hand mocap, full-body mocap, IMU, and hierarchical language annotations for embodied AI, robotics, and world modeling research.

rgbaudiodepthproprioception+3
97K downloadsApr 2026other

Hy-Embodied-0.5-VLA-Data

A large-scale bimanual manipulation dataset with 2,163 hours of high-fidelity demonstrations collected via custom fingertip UMI device with optical motion-capture, spanning 70+ manipulation tasks for training Vision-Language-Action foundation models.

rgbproprioceptionlanguage
96K downloadsJul 2026cc-by-4.0

In-The-Wild Humanoid Dataset

A large-scale dataset of whole-body humanoid robot learning in natural home environments, capturing human teleoperation demonstrations on Unitree G1 across real homes in Southeast Asia with varying layouts, lighting, and clutter.

rgbstereo_irproprioceptionimu+2
77K downloadsJun 2026

In-The-Wild Humanoid Dataset

A large-scale dataset of 500+ hours and 23K+ episodes of whole-body humanoid robot learning captured through human teleoperation demonstrations on Unitree G1 across real homes in Southeast Asia. The dataset includes synchronized visual observations, robot states, actions, and metadata for research on mobile manipulation, bimanual interaction, and long-horizon household skills.

rgbinfraredproprioceptionimu+2
76K downloadsJun 2026