Dataset specification
Dataset specification
Recorded during regular work of our own cleaning company: offices, stairwells, holiday apartments, windows and outdoor areas.
11 data components
- 01
Egocentric video
Operator perspective
- 02
Hand tracking
Position and movement
- 03
Timestamped actions
Precise segmentation
- 04
Task labels
Tasks and single steps
- 05
Object labels
Objects, tools, surfaces
- 06
Object–action relations
What, with which tool, how
- 07
Environment metadata
Room, lighting, conditions
- 08
Success / failure
Incl. mistakes and recoveries
- 09
Before / after state
State around each task
- 10
Multiple workers
Varied movement styles
- 11
Multiple environments
Varied rooms and layouts
Sample capture (illustrative)
Delivered recording with annotations, sensor signals and certification. Values are illustrative.
action: grasp_cloth
object: microfiber_cloth
surface: desk_laminate
hands: L ✓ R ✓ · 21 kp
heart rate: 92 bpm
wrist accel: 0.40 g
Provenance
Site: office, Heide (DE) · Operator ID: W-07 (pseudonymized) · Device: head cam 4K/60 + wrist IMU + chest HR strap
Certification
SHA-256 hash per file · signed capture log · consent ref. C-2026-031 · QC reviewed by 2 annotators
Privacy
No faces or screens in frame · biometric signals stored separately, pseudonymized, only with explicit consent (Art. 9 GDPR)
Diversity & generalization
Workers · Buildings & rooms · Tools & object positions · Lighting, clutter, dirt · Speeds & solution paths · Mistakes & recoveries
Quality control & validation
Quality & completeness · Timestamp sync · Label review · Manual sampling · Flagging unusable clips · Versioning & provenance
Recording workflow
Upload, annotate and certify recordings
Recordings move step by step from upload through review to certification. Each status change is logged with reviewer and comment.
Loading…
Next step
Technical discovery call
Define target tasks, schema and delivery format for your pilot.