DROID (Distributed Robot Interaction Dataset)

A real-robot manipulation dataset collected by 50 people across 564 scenes on three continents using Franka arms.

DROID was collected jointly by 13 institutions including Stanford and UC Berkeley, and released in March 2024. Every site used identical hardware: a 7-DoF Franka Panda arm mounted on a height-adjustable mobile cart, with two third-person ZED 2 stereo cameras and a wrist-mounted ZED Mini, with collectors using a Quest 2 VR controller for teleoperation. Over 12 months, 50 collectors across labs, offices, and homes in North America, Asia, and Europe gathered 76,000 successful trajectories totaling about 350 hours, spanning 564 scenes, 86 task categories, and 1,417 camera viewpoints, along with roughly 16,000 trajectories labeled as failures. It emphasizes “in-the-wild” scene diversity, and models such as π0 include DROID in their pretraining data.

ExampleFor training a policy, one can download the full RLDS-format DROID directly (about 1.7 TB); when high-resolution stereo footage or depth data is needed, use the raw-data release instead (about 8.7 TB).

Also called

DROID

Related

Open X-Embodiment · Real-Robot Data · VR Teleoperation · Data Diversity · RLDS (Reinforcement Learning Datasets) · Franka Emika Panda / Franka Research 3

Sources

DROID 项目主页 (Chinese)
DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset (arXiv 2403.12945)
The DROID Dataset (DROID Docs)

As of

2025-04