Hub.xyz

Multimodal data lab
for physical AI

Hub's mission is to bring embodied AGI closer through real-world data and research.

Become a Contributor Request a Dataset

Frontier labs and top-3 robotics companies run on Hub data.

Network

Global network of
expert contributors

150,000+ active contributors and 730+ commercial sites across 150 countries.

Request a Custom Dataset

live captures · streaming from the field

Pipeline

Frontier datasets,
for frontier labs

One pipeline · Captured · Enriched · Annotated · QA'd

Real-world data

RGB-D

Egocentric human data in commercial and residential environments: calibrated stereo RGB with dense metric depth, synced IMU, 6-DoF VIO, and 3D hand pose.

540,000+hrs captured
10,000+environments

Request Egocentric Sample

RGB+IMU

3D Metric Depth

Dataset 02 / 04

Image

Raw and edited image pairs with pixel-level QA, sourced across the Hub network and refined by our expert community and 100+ professional studios.

2.4Mframes
38verticals

Request Image Sample

Raw photograph — four people, removal target boxed

Raw

Edited

Dataset 03 / 04

Video

Object and hand tracks, temporal action segmentation, and dense scene captions. Frame-accurate for learning physical dynamics.

270kclips
78k hrsduration

Request Video Sample

Dataset 04 / 04

Audio

Multilingual field recordings with speaker diarization, language and accent tags, and per-utterance emotion labels.

80k hrsaudio
47languages
48 kHzsample

Request Audio Sample

ENPull it the moment it goes gold.

ESBaja el fuego antes de que se queme.

Sample Rate 48 kHzBitrate 320 kbpsLanguage EnglishChannels 2ch · stereo

1 / 4

Custom collection

Bespoke collection
in hours, not weeks

Input your hardest data needs. Hub matches commercial environments and individual contributors, enforces diversity caps, and starts delivering within hours.

Request a Datasetend-to-end · you never touch the pipeline