EmbodiedData

VERIFIED DATASET SOURCE DIRECTORY

Where can robotics datasets actually be found?

A curated map of major platforms, research projects and benchmark websites—what each contains, how access works, and whether it is suitable for robot-policy training.

Official projectDiscovery platformAccess may require approval

Three types of source

Understanding the difference prevents confusion and duplicate downloads.

Discovery platform

Hosts or indexes many publishers. Always check the individual dataset card and licence.

Official project

The research team’s authoritative page for documentation, code, data and citation.

Converted copy

The same underlying dataset transformed into LeRobot, RLDS or another format. Verify who produced the conversion.

14 MAJOR SOURCES REVIEWED

Dataset websites and collections

This is a maintained major-source directory, not a claim that every lab repository on the internet is included.

01Discovery platform

Hugging Face Robotics Datasets

Best for
LeRobot datasets, converted datasets and direct file browsing
Formats
LeRobot · Parquet · Video · many others
Access
Usually direct; individual licences apply
Level
Best general starting point
Open source website ↗
02Official collection

LeRobot Dataset Collection

Best for
Datasets ready for the Hugging Face LeRobot ecosystem
Formats
LeRobot
Access
Direct through Hugging Face
Level
Beginner friendly
Open source website ↗
03Official research collection

Open X-Embodiment

Best for
Large-scale cross-robot learning and RT-X research
Formats
RLDS
Access
Project and repository instructions
Level
Advanced
Open source website ↗
04Official dataset project

DROID

Best for
Large, diverse real-world Franka manipulation demonstrations
Formats
RLDS · raw data
Access
Google Cloud download; very large
Level
Advanced
Open source website ↗
05Official dataset project

BridgeData V2

Best for
Language- and goal-conditioned real-world manipulation
Formats
TensorFlow/JAX ecosystem · conversions available
Access
Official project download
Level
Intermediate–advanced
Open source website ↗
06Official benchmark

LIBERO

Best for
Lifelong learning and language-conditioned simulated manipulation
Formats
Benchmark demonstrations
Access
Official download script
Level
Intermediate
Open source website ↗
07Official framework and datasets

robomimic

Best for
Imitation learning and offline learning benchmarks
Formats
HDF5
Access
Dataset download tools and documentation
Level
Intermediate
Open source website ↗
08Official project

ALOHA

Best for
Bimanual teleoperation, ACT and imitation learning
Formats
Project-specific · converted versions exist
Access
Project repository and linked datasets
Level
Intermediate–advanced
Open source website ↗
09Official project

Mobile ALOHA

Best for
Whole-body mobile bimanual manipulation
Formats
Project-specific
Access
Project resources and linked data
Level
Advanced
Open source website ↗
10Publisher collection

NVIDIA Physical AI Datasets

Best for
Simulation and real-world physical-AI releases
Formats
LeRobot and release-specific formats
Access
Individual Hugging Face cards
Level
Mixed difficulty
Open source website ↗
11Dataset catalogue

TensorFlow Datasets — Robotics

Best for
Standardized RLDS loaders and research datasets
Formats
RLDS · TensorFlow Datasets
Access
Programmatic download
Level
Intermediate
Open source website ↗
12Official human-video dataset

Ego4D

Best for
First-person activity understanding and visual pretraining
Formats
Video · annotations
Access
Licence agreement and downloader
Level
Not robot-action data
Open source website ↗
13Official human-video dataset

Ego-Exo4D

Best for
Synchronized first- and third-person skill video with language
Formats
Video · language annotations
Access
Licence approval and downloader
Level
Not robot-action data
Open source website ↗
14General discovery platform

Kaggle Dataset Search

Best for
Community datasets and small experimentation files
Formats
Varies
Access
Kaggle account may be required
Level
Verify quality carefully
Open source website ↗

NOT SURE WHERE TO GO?

Use EmbodiedData first, then open the right source.

Search here for a beginner explanation and requirements. When you are ready, follow the verified link to the publisher for real files, licence terms and current documentation.

Find a dataset →