Skip to main content

LeRobot Studio: The Browser for Embodied AI Data

LeRobot Studio is a browser-based LeRobot dataset visualization tool developed by IO-AI.TECH. It aims to provide an intuitive and efficient data analysis and verification environment for Embodied AI researchers and developers.

Background: The Importance of the LeRobot Dataset Standard​

With the development of the Embodied AI field, high-quality datasets have become a core element. However, robotic data has long suffered from issues such as format fragmentation, low storage efficiency, and difficulty in reuse.

The LeRobot Dataset Standard (v2.1/v3.0) is rapidly becoming the "USB interface" of the Embodied AI field. Top open-source models and simulation platforms, including OpenPi and NVIDIA GR00T (Isaac Lab), already support or are compatible with the LeRobot format. It solves the "Data Tower of Babel" problem that has long plagued the industry through the following features:

  1. Unified Data Structure: Utilizing a combination of Parquet + MP4, it efficiently encapsulates multimodal data such as images, motor states (State), and action commands (Action), solving data compatibility issues across platforms and devices.
  2. Cloud-Native Support: Supports Streaming, allowing data to be loaded directly from cloud storage like Hugging Face Hub or local storage without pre-downloading massive files, significantly improving data access efficiency.
  3. Community Ecosystem Interoperability: The standardized format makes datasets easier to retrieve, index, and share, promoting collaboration and development within the robot learning community.

LeRobot Studio is a universal browser built upon this standard, helping users deeply understand and verify these standardized data resources.

LeRobot Studio Core Features​

LeRobot Studio provides zero-threshold data debugging capabilities, usable without any local environment configuration.

1. Multimodal Data Synchronized Playback​

Robot data includes high-frequency sensor data and video streams with different frame rates. LeRobot Studio implements a timeline alignment mechanism, capable of synchronously displaying:

  • Multi-view camera feeds
  • Robot joint state curves
  • Control command sequences

By dragging the timeline, users can precisely retrace the robot's perception and decision state at any moment.

Multi-view video synchronization

2. State vs. Action Comparative Analysis​

In the process of model training and debugging, analyzing the difference between "Expected Command" and "Actual State" is crucial.

  • Command (Action): Control signals issued by the model or teleoperation device (represented by dashed lines).
  • State (State): Joint feedback actually executed by the robot (represented by solid lines).

Through the Chart Panel, users can intuitively observe latency, jitter, or deviation between the two, thereby quickly locating hardware execution issues or data collection quality problems.

Data curve analysis

3. Rapid Dataset Verification​

LeRobot Studio leverages the local computing power of the browser to support direct loading of local folders or compressed packages. Data does not need to be uploaded to the cloud, ensuring data security and loading speed. At the same time, it also supports remote on-demand loading of cloud datasets via URL, facilitating quick previews.

How to Use​

We provide two convenient access points:

Method 1: IO-AI Official Website​

Access Address: https://io-ai.tech/lerobot/

  • Features: No login required, ready to use.
  • Applicable Scenarios: Quickly viewing local datasets, experiencing official sample data.

Welcome Page

Method 2: IO-AI Data Platform​

Access Address: IO-AI Data Platform

  • Features: Deeply integrated with the data management workflow.
  • Applicable Scenarios: Directly performing quality acceptance and preview within the platform after data collection, annotation, and format conversion.

Interface Guide​

The interface layout of LeRobot Studio is designed specifically for data analysis:

1. Episode List Sidebar​

The left sidebar lists all tasks (Episodes) in the dataset. It supports keyword search and metadata display, helping users quickly filter and locate specific data segments.

Episode Sidebar

2. Raw Data Panel​

Displays the complete JSON data structure of the current frame. This is very useful for developers debugging code, as they can directly copy specific values of timestamp, action, or state for troubleshooting.

Raw Data Panel

3. Playback Control Bar​

Provides frame-level playback control. In addition to standard speed playback, it supports frame-by-frame stepping, facilitating the analysis of instantaneous states during fine manipulation processes such as grasping and assembly.

Playback Control Bar

About Us​

IO-AI.TECH is committed to building efficient infrastructure for Embodied AI. LeRobot Studio is an important part of our data toolchain, designed to improve the efficiency of the data closed loop.