Data & collection
LeRobot dataset
A LeRobot dataset is a robot-learning dataset organised for the LeRobot data model and loaders. LeRobot v3 stores low-dimensional state, action and timestamp fields in Parquet, camera streams in MP4, and schema, task, statistics and episode metadata under a coordinated directory layout.
Also known as: LeRobotDataset, LeRobot data, LeRobot dataset format, LeRobot v3
Updated
What the v3 layout contains
LeRobotDataset v3 uses Parquet for frame-level, low-dimensional fields such as robot state, actions and timestamps. It stores camera streams in MP4 shards and uses metadata to describe feature names, shapes, data types, frame rate, normalisation statistics, tasks and episode offsets.
Version 3 places multiple episodes in larger files instead of requiring one data file and one video file for each episode. The metadata reconstructs episode views from those shared files. The loader can expose individual frames or temporal windows and can stream datasets from the Hugging Face Hub when the dataset and access path support it.
What a LeRobot dataset still needs to explain
The format does not make different robots equivalent. A dataset must still document the robot or human capture setup, action space, units, coordinate frames, camera names, calibration, control rate, episode boundaries and the relationship between observations and actions. A field called action is not meaningful without that context.
A LeRobot export may contain robot commands, human pose, action proxies, simulation state or only observations and annotations. Check the feature schema and dataset card before assuming it is ready for behaviour cloning or deployment on a humanoid.
LeRobot and source recordings
A LeRobot export can be derived from a source such as an MCAP recording, ROS bag or device-native archive. For example, EgoSuite documents both MCAP recordings and LeRobot exports. Keep the source identifier and conversion manifest with the release when later reprocessing, calibration review or annotation changes are possible.
Sources
Related terms
Data & collection
Robot training data
Robot training data is recorded experience used to train, fine-tune, or adapt models for robot perception, prediction, planning, or control. It can include sensor observations, robot state, actions, task instructions, rewards or outcomes, demonstrations, failures, and embodiment metadata. Not every dataset contains every field, but their timing and physical meaning must be clear.
Data & collection
Trajectory
A trajectory is a time-ordered sequence of states or observations, actions and, where applicable, rewards generated as an agent or robot evolves. A complete episode or policy rollout often yields a trajectory, but the terms are not universally identical: trajectories may be partial, while episodes have dataset- or environment-defined boundaries.
Models & learning
Action space
An action space is the set and representation of commands that an agent or robot policy is allowed to choose. In robotics, actions may be discrete choices or continuous values such as joint targets, motor torques, end-effector pose changes, base velocities, or gripper commands. The action space defines what the policy outputs, not necessarily the motion the hardware ultimately executes.
Data & collection
Data synchronisation
Data synchronisation is the process of placing sensor, state, action, annotation, and outcome records on a common timeline so samples that describe the same physical instant or transition can be matched. It requires trustworthy timestamps or trigger relationships and an explicit rule for handling streams with different rates, delays, dropped samples, and clock offsets.
Data & collection
Robot state
Robot state is the set of variables used to describe a robot at a particular time, such as joint positions and velocities, base pose, end-effector pose, gripper state, actuator measurements, or estimated motion. In control theory, a complete state contains enough information to predict future evolution given an action; in robot datasets, “state” often means only the measured or estimated subset that was logged.