Directory Structure
A standard LeRobot dataset directory with two main sections: meta/ and videos/.
LivSyn data collection devices capture large-scale embodied data in real-world operations across retail, dining, offices, homes, automotive services, and more. First-person video and IMU poses are hardware-synchronized and delivered in a unified LeRobot format for direct viewing and download. Ready-made datasets are training-ready with precise semantic annotations, while our in-house data platform enforces strict quality control, compliance, and security for high-quality global delivery.
A standard LeRobot dataset directory with two main sections: meta/ and videos/.
data/chunk-000/episode_000000.parquet stores state, timestamps, and poses frame by frame.
info.json defines features and schema, episodes.jsonl records task lists and frame counts, and tasks.jsonl provides the task dictionary.
Synchronized left-eye and right-eye RGB streams encoded in H.264 and aligned frame by frame.
6DoF head pose in world coordinates, together with 21-joint poses for both hands.
task and task_index provide the task text and index for each frame.
Approximately 30 Hz, with timestamps aligned to capture frame numbers for frame-level traceability.
All samples were captured on site with an EGO headset and preserve complete first-person video and IMU pose data.
Wiping and cleaning automotive glass surfaces, covering side windows and windshields.
LeRobot3.0Sorting, returning, and organizing medical room supplies and work surfaces.
LeRobot3.0Making beds and returning personal items to their proper places in a care ward.
LeRobot3.0Covering 10 industries and 59 real-world locations, with each location broken down into specific tasks and combinable by industry scenario.
We can provide samples, collection plans, and custom data collection support tailored to your industry scenarios and robot embodiment.
Contact Us →