Skip to main content

LeRobot Studio: a browser for embodied AI datasets

About

LeRobot Studio (https://io-ai.tech/lerobot/) is a local LeRobot dataset visualization and inspection tool developed by IO-AI.TECH.
Drag and drop a local LeRobot folder or archive into the browser to play it back—no cloud upload required, for both efficiency and privacy.

Why the LeRobot dataset standard matters​

High-quality datasets are central to embodied AI, yet robot data has long suffered from fragmented formats, inefficient storage, and poor reuse.

The LeRobot dataset standard (v2.1 / v3.0) is becoming a de facto interchange format. Ecosystems such as OpenPI and NVIDIA GR00T (Isaac Lab) already support or interoperate with LeRobot layouts. The standard addresses fragmentation by providing:

  1. A unified structure: Parquet plus MP4 bundles images, motor state, and actions in one coherent schema across platforms and hardware.
  2. Cloud-native access: Streaming lets you load from the Hugging Face Hub or local storage without downloading entire corpora first.
  3. Community interoperability: A common layout improves discovery, indexing, and sharing across the robot-learning community.

LeRobot Studio is a general-purpose viewer built on that standard so you can understand and validate these datasets quickly.

Core features​

LeRobot Studio runs in the browser—no local toolchain required for basic inspection.

1. Synchronized multimodal playback​

Robot logs combine high-rate sensors and video at different frame rates. LeRobot Studio aligns timelines to show together:

  • Multi-camera video
  • Joint state curves
  • Control command sequences

Scrub the timeline to revisit sensing and decisions at any instant.

Multi-camera sync

2. Action vs. state comparison​

During training or teleop debugging, comparing commanded actions to measured state is essential.

  • Action: command from the model or teleop (shown as dashed lines).
  • State: measured joint feedback (solid lines).

The chart panel highlights latency, jitter, or tracking error between the two.

Chart panel

3. Fast validation and the Dataset Health panel​

Processing stays in the browser: load a local folder or archive without uploading to the cloud. You can also open remote datasets by URL for quick previews.

After loading, open the Dataset Health sidebar to see validation results (error / warning / pass), including:

  • Layout: whether meta/info.json, meta/episodes, data/, videos/, etc. match v2.1 or v3.0 rules
  • Meta: required fields such as fps, codebase_version
  • Features: dtype, shape, and related constraints per feature
  • Episodes: episode counts and lengths vs. info, and zero-length segments

Issues are listed explicitly so you can confirm exports quickly—recommended for post-export QA.

How to access it​

Two entry points:

Option 1: IO-AI website​

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

  • No login required.
  • Ideal for local datasets and bundled demos.

Welcome screen

Option 2: IO-AI data platform​

IO-AI data platform

  • Tight integration with data management workflows.
  • Use after capture, labeling, or format conversion for in-platform QA.

UI tour​

1. Episode sidebar​

Lists every episode with search and metadata for fast navigation.

Episode sidebar

2. Raw Data panel​

Shows the full JSON for the current frame—handy for copying timestamp, action, or state values while debugging.

Raw Data panel

3. Playback controls​

Frame-accurate transport including step frame-by-frame for fine-grained inspection (grasping, insertion, etc.).

Playback controls

4. Dataset Health panel​

Open Dataset Health in the sidebar to view validation status (pass / warning / error). Checks run on load and cover v2.1 / v3.0 structure, meta/info, features, and episodes—useful right after export.

About IO-AI.TECH​

IO-AI.TECH builds infrastructure for embodied AI. LeRobot Studio is part of our data toolchain to tighten the data loop.