DROID

Distributed Robot Interaction Dataset — large-scale in-the-wild manipulation data from UT Austin.

Overview

DROID is a large-scale manipulation dataset collected by 50 human operators across 564 real-world scenes. It emphasizes diversity of environments, objects, and tasks for robust imitation learning and policy transfer.

Scale & Content

76,000+ trajectories
350 hours of interaction
86 tasks (pick-place, assembly, tool use, etc.)
564 scenes, 50 collectors
RGB-D, proprioception, actions

Official Links

droid-dataset.github.io — Project site
github.com/droid-dataset/droid — Code & data
Hugging Face: lerobot/droid_100 — LeRobot format

Citation

See the project site for BibTeX and paper links (RSS 2024).