CLARU
oxe-auge2025cc-by-4.0

language_table_train_5000_10000_augmented

An augmented dataset of 5,000 episodes containing language-instructed robotic manipulation tasks with cross-embodiment augmentation for 8 different robot platforms. Includes visual observations, joint states, end-effector poses, and natural language instructions.

Downloads82
Episodes5000

Why This Matters for Physical AI

This dataset enables training of cross-embodiment robotic policies by providing augmented visual and proprioceptive observations paired with natural language instructions across multiple robot platforms, advancing generalization in language-conditioned robot learning.

Technical Profile

Modalities
rgbproprioceptionlanguage
Robot Embodiments
Google RobotJacoKinova3KUKA IIWAFranka PandaSawyerUR5e
Environment
lab
Task Types
manipulationlanguage-instructed manipulation
Episodes
5000
Data Format
parquet
Annotation Types
language_instructions
License
cc-by-4.0
Part of the OXE-AugE family

Access

View on HuggingFaceRead Paper

Need custom rgb data?

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

Request a Sample Pack

Related Datasets

language_table_train_80000_85000_augmented

An augmented subset of the Language Table dataset containing 5,000 episodes with cross-embodiment visual and proprioceptive data for 8 different robot platforms, augmented through domain randomization techniques to enable scaling of cross-embodiment policy learning.

rgbproprioceptionlanguage
86 downloadsNov 2025cc-by-4.0

language_table_train_180000_185000_augmented

An augmented subset of the Language Table dataset containing 5,000 episodes of robot manipulation data with language instructions, augmented to include cross-embodiment trajectories for 8 different robot platforms.

rgbproprioceptionlanguage
86 downloadsNov 2025cc-by-4.0

language_table_train_45000_50000_augmented

An augmented subset of the Language Table dataset containing 5,000 episodes with cross-embodiment robot augmentation, providing multi-robot versions of trajectories with natural language instructions.

rgbproprioceptionlanguage
85 downloadsNov 2025cc-by-4.0

language_table_train_175000_180000_augmented

An augmented subset of the Language Table dataset containing 5,000 episodes with multi-robot visual and proprioceptive observations paired with natural language instructions, generated through cross-embodiment augmentation across 8 robot platforms.

rgbproprioceptionlanguage
84 downloadsNov 2025cc-by-4.0

language_table_train_25000_30000_augmented

An augmented subset of the Language Table dataset containing 5,000 episodes with 80,044 frames of multi-robot manipulation data with language instructions. The dataset provides cross-embodiment robot trajectories with augmented observations for 8 different robot platforms.

rgbproprioceptionlanguage
84 downloadsNov 2025cc-by-4.0

language_table_train_95000_100000_augmented

An augmented subset of the Language Table dataset containing 5,000 episodes with cross-embodiment image augmentation for 8 different robot platforms, enabling multi-robot policy learning from language instructions.

rgbproprioceptionlanguage
83 downloadsNov 2025cc-by-4.0