AvailablePhysical AIGR-EGO-02

Ego-Exo Workplace Tasks

Synchronized first-person and multi-view video of real work in factories, hotels and offices.

Request sample Download spec (PDF)Get pricing

Hours of video: 4,300+
Synchronized views: 4
Frame rate: 30 fps
Hours per participant: ≤ 10

Illustrative rendering of annotation layers.

01Overview

Real workers doing their real jobs — small-part assembly and packaging on factory floors, housekeeping in hotel rooms, tidying and cleaning in offices and reception areas — recorded by a head-mounted camera and three fixed cameras at the same time. Sessions are synchronized and calibrated. Annotators describe each action in free text with timestamps, and a pose model estimates body keypoints; raw footage can also be licensed without annotations.

Modalities

Egocentric video
3 synchronized exo views
Camera calibration
Action segments
Model-estimated body keypoints

Use cases

Pre-training vision-language-action (VLA) and humanoid policies on real work
Ego-exo correspondence and view-invariant representation learning
Procedure, action and task recognition in workplaces
Human pose and motion modelling from calibrated multi-view video

02Specification

Capture

Head-mounted camera: Wide-angle action camera, both hands in view during manipulation
Fixed cameras: 3 static cameras: one full-body view at 2–3 m, one frontal view within 2 m, one on the hands
Resolution: At least 2 MP (1920 × 1080) per camera
Frame rate: 30 fps
Synchronization: Dynamic QR code shown to every camera at the start of each session
Calibration: Calibration-board intrinsics and extrinsics per session

Coverage

Manufacturing: ≈ 78% — assembly, packaging, sorting, labelling and inspection
Hospitality: ≈ 13% — bed making, cleaning and tidying in hotel rooms and lobbies
Offices and public spaces: Meeting rooms, reception, offices, lounges and dining areas
Participants: Real workers in their own workplaces; at most 10 hours per person
Content rules: No staged scenes or actors; no idle walking; tasks occur naturally
Task concentration: Repetitive bench work dominates; the largest single tasks run 300–450 hours each. Per-task hours on request

Annotation

Action segments: Timestamped free-text descriptions written by annotators in Chinese; task names are not normalized to a fixed taxonomy. English translation on request
Body keypoints: 17-point (COCO) body keypoints per fixed camera, estimated by a pose model with per-point confidence
3D pose: Triangulated from the calibrated fixed cameras; relative scale, metric conversion in preparation
Hand keypoints: Available as a custom add-on
Acceptance review: Per-session checks of sync, camera coverage, image quality and content

Compliance

Consent: Written participant consent covering AI training
Privacy: Face and screen blurring available before delivery
Provenance: Unique session ID with capture metadata and acceptance record
Delivery options: Raw synchronized video with calibration, with or without annotations

03Schema

Field	Type	Description
session_id	string	Unique capture session identifier.
ego	mp4	Head-mounted video.
exo_cam1 … exo_cam3	mp4	Synchronized fixed-camera videos.
calibration	yaml	Per-camera intrinsics and extrinsics.
keypoints_2d	json	Per-camera, per-frame 17-point body keypoints with model confidence.
keypoints_3d	json	Per-frame triangulated 17-point body keypoints (relative scale).
actions	json	Action segments: start, end and description.
metadata	json	Scene, task, devices, duration and acceptance status.

04File structure

Directory layout

ego-exo-workplace/            # indicative; final layout confirmed per delivery
├── manifest.csv
└── sessions/<scene>/<session_id>/
    ├── ego/
    ├── exo_cam1/
    ├── exo_cam2/
    ├── exo_cam3/
    ├── calibration.yaml
    ├── keypoints_2d.json
    ├── keypoints_3d.json
    ├── actions.json
    └── metadata.json

05Sample

Sample on request.

Raw multi-view sessions from manufacturing and hospitality scenes
Calibration files, action segments and body keypoints

Size: On request
Formats: MP4, YAML, JSON

Request sample Download spec (PDF)

06License

Graviti Commercial Data License

Licensed per hour of footage, with rights to train and commercially deploy models. Custom collection to your task list is available.

How we license data →

07FAQ

Questions about Ego-Exo Workplace Tasks

What is the Ego-Exo Workplace Tasks dataset?

Ego-Exo Workplace Tasks (GR-EGO-02) is a Physical AI training dataset from Graviti. Real workers doing their real jobs — small-part assembly and packaging on factory floors, housekeeping in hotel rooms, tidying and cleaning in offices and reception areas — recorded by a head-mounted camera and three fixed cameras at the same time. Sessions are synchronized and calibrated. Annotators describe each action in free text with timestamps, and a pose model estimates body keypoints; raw footage can also be licensed without annotations.

What is Ego-Exo Workplace Tasks used for?

Typical uses: pre-training vision-language-action (vla) and humanoid policies on real work, ego-exo correspondence and view-invariant representation learning, procedure, action and task recognition in workplaces and human pose and motion modelling from calibrated multi-view video.

What modalities and annotations does Ego-Exo Workplace Tasks include?

Ego-Exo Workplace Tasks includes Egocentric video, 3 synchronized exo views, Camera calibration, Action segments and Model-estimated body keypoints. Its schema has the fields session_id, ego, exo_cam1 … exo_cam3, calibration, keypoints_2d, keypoints_3d, actions and metadata.

How can I get a sample of Ego-Exo Workplace Tasks?

Ego-Exo Workplace Tasks is available. Pilot data is available on request at graviti.com/datasets/ego-exo-workplace/sample.

How is Ego-Exo Workplace Tasks licensed?

Graviti Commercial Data License. Licensed per hour of footage, with rights to train and commercially deploy models. Custom collection to your task list is available.

Related datasets

GR-EGO-01 · Physical AIAvailable

Egocentric Household Activities

First-person video of real people doing real chores in real homes.

Hours of video: 10,000+
Contributors: 2,400+

View spec

GR-TEL-01 · Physical AIEarly access

Bimanual Teleoperation Trajectories

Synchronized multi-view robot demonstrations with language instructions.

Episodes: 120,000+
Task families: 180

View spec

Get started

Need more than Ego-Exo Workplace Tasks?

We extend catalog datasets with new environments, tasks, languages and annotation layers — or collect from scratch to your spec.

Contact sales Browse datasets