Egocentric Videos
Test Blog Post
Short summary of first blogpost
Jan 12, 2026
Real World. Real Intent. Raw Data.
TrainRobotswith
EgocentricHuman Data
High-quality, intent-rich POV datasets to fuel the next generation of Embodied AI.
The Problem
Roboticsisstarved
of physical reality.
Training robots for the real world requires real-world data. Capturing authentic human behavior at scale has been the missing piece.
01
Lab demos are staged
Controlled environments miss real-world messiness—clutter, varied lighting, unexpected obstacles.
02
Teleoperation doesn't scale
Manual data collection is slow and expensive. Hours of operator time for minutes of useful data.
03
Sim2Real gap persists
Simulated environments fail to capture household physics and the unpredictability of real homes.
04
Static datasets go stale
One-time data releases can't keep up. Robots need continuous, fresh data for ongoing learning.
Our Solution
Observationallearning
at scale.
Phone-first egocentric capture from real homes—focused on goal-driven, intent-rich tasks.
01
Ego-view capture
First-person perspective mirrors how robots will perceive the world—not third-party observation.
02
Household reality
Real-world messiness: clutter, varied lighting, unexpected layouts. Non-staged, diverse homes.
03
Intent over micro-motion
Long-horizon, goal-driven tasks. What to achieve matters more than exact trajectories.
04
Continuous collection
Phone-first capture scales effortlessly. Lightweight mounts, no expensive hardware.
Egocentric Capture
First-person POV
LIVE
kitchen_unload_001.mp400:47
Annotated Output
Verb + Noun labels
0.0sopendoor
2.3spick_upglass
4.1splaceshelf
Our Approach
Macro tasks,micro annotations.
Goal-level task packs segmented into atomic actions with precise timestamps.
Goal-drivenFrame-levelMachine-ready
Raw Capture
00:47
glass
shelf
pick_up → glass
Annotated
12 actions
opendoor→
pick_upglass→
placeshelf→
closedoor
Our Pipeline
Fromcapture
to delivery.
Quality at every step while scaling collection
across our global contributor network.
01
Calibrate
One-time setup test ensures proper mount and framing.
02
Capture
Contributors record household tasks from egocentric view.
03
Ingest
Secure upload via presigned URLs. Zero server contact.
04
QC
Automated + manual review for technical and task quality.
05
Annotate
Verb+Noun segmentation with precise timestamps.
06
Deliver
Versioned dataset packs for your training pipeline.
Annotation standard:Verb + Noun
Every video segmented into atomic actions following EPIC-KITCHENS vocabulary. Precise timestamps, machine-readable labels.
open+doorpick_up+glassplace+shelfPilot in progress
What We Collect
Pilottasklibrary,
expanding.
15-20 core household tasks focused on activities
robots need to master for real-world deployment.
🍳8 tasks
Kitchen
- Dishwasher load/unload
- Table setting/clearing
- Food prep
- Appliance use
👕6 tasks
Laundry
- Clothes folding
- Laundry sorting
- Machine operation
- Hanging clothes
📦5 tasks
Organization
- Cabinet organizing
- Shelf arrangement
- Drawer sorting
- Item placement
🚶4 tasks
Navigation
- Room-to-room
- Door opening
- Carrying objects
- Path planning
0
Tasks Defined
0
Categories
∞
Scalable
Explore task library
Login required
Data Schema
EgosenseHome
Taxonomy.
Built on EPIC-KITCHENS Verb-Noun schema,
extended for whole-home manipulation tasks.
0+
Verbs
0+
Nouns
0+
Categories
Key Features
- Atomic Actions (Verb + Noun)
- Training-grade temporal grounding
- Household object registry: tools, containers, surfaces
- Mapping-ready for RT-X and Ego4D
Verb Categories
Generic15Kitchen20Laundry12Cleaning10Organization13
egosense_dataset.json
v1.0.1
// dishwasher_unload.json
{
"task_id": "TASK_0042",
"task": "dishwasher_unload",
"taxonomy_version": "1.0.1",
"contributor_id": "TR_0015",
"duration_ms": 47200,
"fps": 30,
"resolution": "1920x1080",
"actions": [
{
"action_id": 1,
"start_ms": 0, "end_ms": 2300,
"verb": "open", "verb_class": 12,
"noun": "dishwasher_door", "noun_class": 45
},
{
"action_id": 2,
"start_ms": 2300, "end_ms": 4100,
"verb": "pick_up", "verb_class": 5,
"noun": "glass", "noun_class": 23
},
{
"action_id": 3,
"verb": "place", "noun": "cabinet_shelf"
}
],
"metadata": {
"environment": "kitchen",
"lighting": "natural",
"device": "iPhone 14 Pro"
}
}
// laundry_folding.json
{
"task_id": "TASK_0089",
"task": "laundry_folding",
"taxonomy_version": "1.0.1",
"duration_ms": 62400,
"actions": [
{
"verb": "pick_up", "noun": "t-shirt",
"confidence": 0.94
},
{
"verb": "fold", "noun": "t-shirt",
"confidence": 0.97
},
{
"verb": "put", "noun": "drawer"
}
]
}
// table_wiping.json
{
"task": "table_wiping",
"region": "TR",
"actions": [
{ "verb": "pick_up", "noun": "cloth" },
{ "verb": "wipe", "noun": "table" },
{ "verb": "rinse", "noun": "cloth" }
]
}
// cabinet_organizing.json
{
"task": "cabinet_organizing",
"object_count": 12,
"actions": [
{ "verb": "open", "noun": "cabinet_door" },
{ "verb": "remove", "noun": "plate" },
{ "verb": "stack", "noun": "plate" },
{ "verb": "put", "noun": "shelf" }
]
}
// dishwasher_unload.json
{
"task_id": "TASK_0042",
"task": "dishwasher_unload",
"taxonomy_version": "1.0.1",
"contributor_id": "TR_0015",
"duration_ms": 47200,
"fps": 30,
"resolution": "1920x1080",
"actions": [
{
"action_id": 1,
"start_ms": 0, "end_ms": 2300,
"verb": "open", "verb_class": 12,
"noun": "dishwasher_door", "noun_class": 45
},
{
"action_id": 2,
"start_ms": 2300, "end_ms": 4100,
"verb": "pick_up", "verb_class": 5,
"noun": "glass", "noun_class": 23
},
{
"action_id": 3,
"verb": "place", "noun": "cabinet_shelf"
}
],
"metadata": {
"environment": "kitchen",
"lighting": "natural",
"device": "iPhone 14 Pro"
}
}
// laundry_folding.json
{
"task_id": "TASK_0089",
"task": "laundry_folding",
"taxonomy_version": "1.0.1",
"duration_ms": 62400,
"actions": [
{
"verb": "pick_up", "noun": "t-shirt",
"confidence": 0.94
},
{
"verb": "fold", "noun": "t-shirt",
"confidence": 0.97
},
{
"verb": "put", "noun": "drawer"
}
]
}
// table_wiping.json
{
"task": "table_wiping",
"region": "TR",
"actions": [
{ "verb": "pick_up", "noun": "cloth" },
{ "verb": "wipe", "noun": "table" },
{ "verb": "rinse", "noun": "cloth" }
]
}
// cabinet_organizing.json
{
"task": "cabinet_organizing",
"object_count": 12,
"actions": [
{ "verb": "open", "noun": "cabinet_door" },
{ "verb": "remove", "noun": "plate" },
{ "verb": "stack", "noun": "plate" },
{ "verb": "put", "noun": "shelf" }
]
}
Mapping-ready for RT-X, Ego4D, and VLA pipelines
Compatible with
🍳EPIC-KITCHENS🤖Open X-Embodiment👁️Ego4D🧠VLA Models
Delivery
What's in thepack?
Everything to integrate into your pipeline. Originals + standardized encodes included.
Original source recordings preserved in full quality. Standardized encodes available for training compatibility: 1080p, 30fps, H.264 codec. Consistent aspect ratios across all videos.
MP4 formatOriginals preservedStandardized encodes
Regional Focus
Generalizationthrough
diversity.
Different home styles, clutter levels, and cultural patterns.
Robust training data for models that generalize.
9
Countries
Active
Planned
Active Regions
🇹🇷
Turkey
50+
🇬🇧
UK
Q2
🇦🇪
UAE
Q2
🌐
6+ More
2025
Why Diversity Matters
01
Varied layouts
Different floor plans, room sizes, and furniture arrangements.
02
Lighting conditions
Natural and artificial light across times of day and seasons.
03
Real clutter
Non-staged homes with authentic object placement and density.
Docs
API Reference
Schema docs, type definitions, and field specs
Download
Sample Data
Preview datasets with full annotations
Code
Integration
PyTorch loaders, training scripts, examples
Papers
Research
Methodology papers and benchmark results
Contact
Let's talk about your
data needs.
Research datasets or custom collection solutions—we'd love to hear from you.
hello@egosense.ai
Response within 24h
What happens next
We review your inquiry
Discovery call
Sample data access
Custom proposal
Hiring Contributors
Earn money recording everyday tasks from home