Embodied AI Dataset

Industrial Assembly and Maintenance Datasets

Creator: Changsha Langhui Information Technology Co., Ltd.
Keywords: ,Embodied-intelligence datasets,Robot imitation learning,World-model training,Langhui,AItraining data

15,000+episodes . Multi-Specific 3D Video . Tool position versus twist curve . Maintenance of SOP text alignment

episodes15,000+

Strength sampling1kHz

Tool Track6D position

Automobile/3C/home electricity/wire beams4 Large industries

Process Text CorrespondSOP Alignment

Dataset Overview

Definition：Industrial assembly and maintenance data sets are smart data sets provided by Long Salangye Information Technology Ltd. The screws, plugs, cable sorting and equipment spot-checking missions under real plant scenarios provide multi-perspective stereo video, tool position positions, twist curves and maintenance of SOP text matching to support industrial quality and predictive maintenance models.

Dataset Name	Industrial Assembly and Maintenance Datasets
Total Data Volume	15,000+ episodes
Data Type	Steeo RGB, depth map, power/twirl, tool 6D position, SOP text
Video Specifications	1920×1080 @ 30 FPS
Delivery Format	LeRobot 3.0 / RLDS / HDF5 / JSON Metadata
License Type	Licensed Data or Custom Projects

Core Data Fields

Each piece of data contains complete structured fields, covering all requirements from basic metadata to professional analysis.

Field Category	Typical Fields
Multi-Perspective Stereo Video	Wings and lefts RGB, depth map, camera in-sync/extras, time stamp, factory site marking, light conditions
Tools and Workspaces	6D position, tool type, grab point, contact status, screw/plugin specification, assembly order index
Hard sense and twist curve	Twisting sensor time series, peak, twist angle, gliding/writing mark, exposure threshold
Maintain SOP and Mass Check	Standard operating procedures, quality test results, abnormal labels, maintenance cycle, engineer review records, video alignment of time axes
Metadata and Compliance	Production line number, equipment model, process version, authorization range, data safety level

Data Sample Showcase

The following is a de-identified structured JSON sample illustrating the typical data format.

Other Organiser
Episode id: "IND-2026-0045231",
"state": "assembly line A3",
"task": "screw m6 panel",
"video": "left": "L 0045231.mp4", "right": "R 0045231.mp4", "depth": "D 0045231.mkv" ,
"tool pose": [x, y, z, qx, qy, qz, qw], "gripper state": [0,1,1,1],
"torque curve": [0.0, 0.05, ...], "torque nm": [0.2, 0.8, 2.1, 2.4], "angle deg": [0, 15, 90, 180],
"sop step": 3,
"sop text": "Use the twister to screw M6 to 2.5 N.m."
"Quality": "ok": true, "defect": full, "inspector": "auto"
♪ I'm sorry ♪

AI Application Scenarios

Industrial mechanical arm imitation learning

Multi-perspective video and power data training for robot-to-end strategy.

Predictive maintenance

Automation of industrial quality checks

Digital twin imitation

Real SOP data-driven plant line digital twinning and process optimization.

Tool Use Understanding

The tool position trajectory supports robotic tool selection and operational planning.

Cross-factor migration studies

Multi-position, multi-machine data enhancement models have a broad capability in new production lines.

Frequently asked questions

What industry scenarios?

The project covers real work places such as auto manufacturing, 3C electronics, domestic electrical assembly, cable beams, equipment spot-checking and maintenance.

How's the power data accuracy?

Twisted sensor sampling rate 1 kHz, accuracy > 0.5% FS, and provides peaks synchronized with angles.

Provide SOP text?

Standard operating step text and alignment with video time axes to support visual-linguistic-action joint training.

How to Get a Data Sample?

Please contact the Changsha Langhui Information Technology Co., Ltd. DataAssetsAPI Platform.