LeRobot Dataset Logger

SUMMARY

LeRobotDatasetLogger records episodic robot demonstrations into writable LeRobot datasets. It manages dataset creation and resumption, episode recording, frame logging, and dataset finalization.

Install Telekinesis Skill Library

Create an API key, install telekinesis-ai and module specific installations and dependencies in a Python environment.

Open installation →

Install

Install the LeRobot dependencies for the Data Engine:

pip install "telekinesis-ai[dataengine-lerobot]"

When to Use LeRobotDatasetLogger

Use the logger to:

• Record teleoperated demonstrations
• Collect imitation learning demonstrations
• Capture observations and actions from a real robot
• Collect simulation rollouts
• Continue data collection in an existing LeRobot dataset

To inspect, analyse or process an existing dataset, use LeRobotDataset instead.

What does LeRobotDatasetLogger provide?

LeRobotDatasetLogger supports two recording workflows:

Workflow	Description
Record	Create a dataset and record episodes containing feature-complete frames.
Resume	Reopen an existing dataset and continue recording new episodes.

The resulting LeRobotDataset contains a sequence of episodes, where each episode contains a sequence of frames recorded at the configured dataset frequency.

For more details on the dataset representation, see LeRobotDataset.

How Does Logging Workflow Work?

LeRobotDatasetLogger manages the recording lifecycle while your application supplies the observations and actions to be logged.

A recording session typically follows this flow:

Observation

Camera• RGB• RGB-D

Robot State• Joint State• Gripper State• End-Effector State

Action

Robot Command• Joint Position Targets• Velocity Targets• Cartesian Pose• Gripper Command

LeRobotDatasetLogger

1 · Instantiate logger

createoverwriteresume

2 · Start episode

start_episode()

3 · Log frames

log(frame)

4 · Stop or discard episode

stop_episode()discard_episode()

5 · Close logger

close()

LeRobot Dataset

Episode 0├─ Frame 0├─ Frame 1└─ ...

Episode 1└─ ...

The logger can create, overwrite, or resume a dataset, then records frames into an active episode. To learn how to use these modes, see the record workflow class reference and resume workflow class reference.

A completed episode is persisted with stop_episode(), while interrupted or invalid episodes can be removed with discard_episode(). When data collection is complete, close() finalizes the logger.

To learn how to represent observations and actions, see the LeRobot Dataset overview and Feature Schema.

Guides

Select a workflow below to create and record a new dataset or append episodes to an existing dataset.

Record a LeRobot Dataset

Create a writable LeRobot dataset and record robot demonstration episodes frame by frame.

Start recording →

Resume LeRobot Recording

Reopen an existing LeRobot dataset and append new episodes without redefining its schema.

Continue recording →