Back to home

Browse Datasets

Every dataset here has video.

Only one modality can't be captured by a camera: force. That's what we build hardware for.

Flagship, only from BLO LAB
Force / TactileWaitlist · First release Q4 2026Collected in-house
BLO-TOUCH-100

Blomega Touch-100

100–1,000 hours of synchronized glove force + bend + IMU + egocentric video across the four household work loops we run today: peeling, kneading, folding, washing. One clock, one schema, real kitchens.

Collected under informed consent · GDPR-compliant · commercial-use licensing
Channels · hardware-synced @ 100 Hz
  • 100–1,000 h · episode-segmented
  • 10 bend channels + 3 fingertip force pads
  • 6-axis IMU · wrist
  • 1080p POV video · 30 fps
  • 4 task families · peel / knead / fold / wash
Force / TactileWaitlist · First release Q1 2027Collected in-house
BLO-NEWTON-1K

Newton-1K

A force-calibrated grasp corpus: 1,000 object–grip pairs captured against reference load cells, so a policy learns absolute newtons, grasp onset, slip, regrip, release, not just relative pressure curves.

Collected under informed consent · GDPR-compliant · commercial-use licensing
Channels · hardware-synced @ 100 Hz
  • 1,000 object–grip pairs
  • Load-cell ground truth · per trial
  • Grasp / slip / regrip / release labels
  • Object mass, material + geometry metadata
  • Force + bend + IMU + POV video @ 100 Hz
Force / TactileWaitlist · First release Q4 2026Collected in-house
BLO-TOUCH

Contact-Rich Manipulation

Force + finger pose + video, hardware-synchronized at 100 Hz inside real work tasks, kitchen, workshop, hardware assembly. The single modality a camera cannot see, captured at the point of contact by GX-1.

Collected under informed consent · GDPR-compliant · commercial-use licensing
Channels · hardware-synced @ 100 Hz
  • 10 bend channels · per-finger
  • Fingertip force pads · thumb + index + middle
  • 6-axis IMU · wrist
  • 1080p egocentric video · 30 fps
  • Object + task labels · per episode
Embodied AI, video & motion
PerceptionLicensed catalog
BLO-MULTIVIEW

Multi-Camera / Multi-Angle

Synchronized views of the same scene for geometric and spatial learning.

Licensed catalog · consent verified upstream · commercial use
· Physics & motion· depth estimation· 3D reconstruction
License:BLO-DLA-1.0
Locale:multi
Task:3D reconstruction
Format:MP4, JSON
Size:412 GB
ManipulationLicensed catalog
BLO-HOI

Human-Object Interaction

People manipulating and interacting with objects and environments across real-world contexts.

Licensed catalog · consent verified upstream · commercial use
· Embodied AI· action recognition· physical reasoning
License:BLO-DLA-1.0
Locale:multi
Task:HOI recognition
Format:MP4, MCAP
Size:2.1 TB
EgocentricLicensed catalog
BLO-EGO

Egocentric

First-person POV footage of how people see, navigate, and interact with the real world.

Licensed catalog · consent verified upstream · commercial use
· Embodied AI· real-world perception· first-person
License:BLO-DLA-1.0
Locale:multi
Task:egocentric perception
Format:MP4, IMU
Size:1.8 TB
Also from Blomega

Content and speech catalogs from the Blomega DataOps side of the house, separate business, shared team. Full portal at blomega.com.

Video & face, content catalog
VideoLicensed catalog
BLO-SCRIPT

Screenplay-Aligned

Scene-level script-to-video alignment across genres, episodes, films, raw footage, and events.

Licensed catalog · consent verified upstream · commercial use
· Text-to-video· narrative coherence· visual reasoning
License:BLO-DLA-1.0
Locale:en
Task:text-to-video
Format:MP4, TSV
Size:890 GB
FaceLicensed catalog
BLO-FACE

Talking Heads

Clear facial movement and articulation across demographics for expressive face generation.

Licensed catalog · consent verified upstream · commercial use
· Avatar & digital humans· lip-sync· expression modeling
License:BLO-DLA-1.0
Locale:multi
Task:avatar generation
Format:MP4, JSON
Size:310 GB