EmbodiedHub LogoEmbodiedHub
Back to Search
D

DROID

DatasetActive

DROID (Distributed Robot Interaction Dataset) is a large-scale real-world robot manipulation dataset by Khazatsky, Pertsch, Levine, Finn and 80+ co-authors. Contains 87k+ trajectories, 350+ hours, across 564 scenes and 86 tasks. Features standardized hardware/software across multiple labs, multi-camera observations, and natural language instructions. The most important real-world robot manipulation dataset for training generalist policies.

Details

Updated:6/20/2026
https://droid-dataset.github.io
sample count87000
licenseCreative Commons Attribution 4.0
modalityvision, proprioception, actions, language

Tags

in-the-wildreal-worldmulti-scenelanguage-conditionedlarge-scalerobot-manipulationbenchmarkgeneralization

Relationships

π0 (pi-zero)
Model
Used to Train
OpenVLA
Model
Used to Train
ALOHA Dataset
Dataset
RELATED_TO
LeRobot
Project
USED_BY

Sources

https://droid-dataset.github.io
website
Visit

Appears In

Embodied AI Landscape

An entry-point knowledge page providing an overview of the embodied AI ecosystem, including companies, robots, models, datasets, and open-source tools.

Embodied AI Datasets

A collection of key datasets used for training embodied AI models, including robot manipulation and vision-language-action models.