Robot data format guides

HDF5 robot data is useful for large structured arrays, trajectories, pose, sensor streams, and compact robot episodes. Define schema version, groups for observations/actions, timestamps, task labels, and metadata attributes before reviewing samples so you can verify that delivery matches the training pipeline.

• HDF5 robot data robotics dataset
• HDF5 robot data training data

LeRobot format is useful for developer-friendly robot learning datasets and policy training pipelines. Define episode metadata, observation tensors, action tensors, timestamps, and repo-compatible manifest before reviewing samples so you can verify that delivery matches the training pipeline.

• LeRobot format robotics dataset
• LeRobot format training data

MCAP is useful for time-synchronized robotics logs, compressed video topics, IMU, state, and action messages. Define topic schema, timestamp domain, compression settings, camera topic, state topic, and manifest before reviewing samples so you can verify that delivery matches the training pipeline.

• MCAP robotics dataset
• MCAP training data

Parquet robot data is useful for large robotics dataset hubs, frame tables, episode metadata, sharded time-series records, and LeRobot-compatible distribution. Define episode index, frame offsets, task descriptions, feature schema, video references, statistics, and split metadata before reviewing samples so you can verify that delivery matches the training pipeline.

• Parquet robot data robotics dataset
• Parquet robot data training data

Pickle robot data is useful for Python-first benchmark releases that package demonstrations, robot state dictionaries, observations, and task metadata. Define schema documentation, Python version notes, object keys, observation/action fields, conversion script, and checksum manifest before reviewing samples so you can verify that delivery matches the training pipeline.

• Pickle robot data robotics dataset
• Pickle robot data training data

Point cloud is useful for 3D scene geometry, object reconstruction, LiDAR/depth capture, navigation perception, and manipulation planning. Define coordinate frame, units, sensor intrinsics and extrinsics, timestamps, segmentation or object labels where available, and source RGB/depth references before reviewing samples so you can verify that delivery matches the training pipeline.

• Point cloud robotics dataset
• Point cloud training data

RLDS is useful for reinforcement learning and robotics episodes with observations, actions, rewards, and metadata. Define episode ID, observation stream, action stream, timestamps, task label, and success flag before reviewing samples so you can verify that delivery matches the training pipeline.

• RLDS robotics dataset
• RLDS training data

ROS bag is useful for robot-native data collection where buyers need ROS topics preserved for replay or conversion. Define topic list, message types, timestamps, sensor calibration, and conversion notes before reviewing samples so you can verify that delivery matches the training pipeline.

• ROS bag robotics dataset
• ROS bag training data