Home · Open data · Manipulation · Robot video

Language-Table

2,701 hours of raw xArm6 teleoperation behind 442k real language-labelled block-pushing trajectories, plus simulated splits; RLDS on a public GCS bucket.

Published by Google Robotics, 2022-10

An xArm6 pushing coloured wooden blocks on a light tabletop
Open data

Frame from the Language-Table dataset, Google Robotics, Apache 2.0 (CC BY 4.0 via Open X-Embodiment).

Commercial use OK

Apache 2.0 (CC BY 4.0 via Open X-Embodiment)

PriceFree

The publisher's licence permits commercial use and redistribution, with attribution. Repository is Apache 2.0; the same data ships under CC BY 4.0 inside Open X-Embodiment. Both permit commercial use.

Get dataset Opens github.com in a new tab
From a terminal
gsutil -m cp -r gs://gresearch/robotics/language_table/0.0.1 .

Nine separate TFDS directories: language_table (real, 442,226 episodes), language_table_sim (181,020), blocktoblock_sim (8,000), blocktoblock_4block_sim (8,298) and five oracle sim sets of 200,000 each. Bucket is public. Total size not published; the real split is large. Also mirrored inside Open X-Embodiment as language_table.

Specification, as published

Hours (stated)
2,701 h
Episodes (stated)
442,226 episodes
Scale
nearly 600,000 language-labeled trajectories
Task
Pick & place, Sorting & packing, Other
Environment
Indoor – lab
Modality
RGB, Actions, Language, Proprioception
Capture
CCTV Scene
In frame
Robot hands
Captured in
North America
Embodiment
UFACTORY xArm6 with cylindrical pusher end-effector (real and simulated)
Frame rate
5 fps
Resolution
640x360
Formats
RLDS/TFRecord
Hosted on
Google Cloud Storage

Figures are the publisher's own. Kinetic Blocks has not measured this dataset.

About this dataset

Language-Table is Google's dataset of language-conditioned tabletop block rearrangement with an xArm6 constrained to 2D pushing. The real-robot split holds 442,226 hindsight-relabelled trajectories drawn from about 2,700 hours of raw teleoperation, and eight simulated splits add human and scripted oracle data. Observations are RGB frames from a fixed RealSense camera plus effector translation actions and natural-language instructions, stored as TFDS/RLDS shards in the gresearch GCS bucket.

Publisher's own task names: block to block, block to block relative, block to absolute location, block to relative location, separate blocks, open-vocabulary language instructions.

What Kinetic Blocks did, and did not do

Indexed and linked from the publisher. Not hosted, verified or graded by Kinetic Blocks. The publisher's terms govern. The licence shown is the one stated on the publisher's page, linked above. Nothing on this page is a claim by Kinetic Blocks.

Want the verified ones too?

Request access to the marketplace, where every paid listing is verified against its files and graded before it goes on sale.