The world, through their eyes.
First-person footage of real people doing real work, synchronized with audio and motion telemetry and labelled with actions and objects for physical AI.
Available as- 01 · Dataset
- 200K+ hrs synced video, IMU, action labels
- 02 · Environment
- Action and manipulation benchmarks
Licensed, stocked, ready to ship.
Plus custom collection for what doesn't exist yet.
Egocentric video and activity data
Processed across physical AI projects
Real-world scenes analysed
Accelerometer and gyroscope per session
Head-mounted manufacturing sessions
Continuous monitoring and review
Footage built to train on.
Task diversity, viewpoint diversity and failure cases included on purpose, because a dataset of only successful grasps teaches less than one that shows where grip slips.
- Head-mounted video
- First-person footage of manipulation and workflow tasks in real environments.
- Synchronized audio
- Ambient and spoken audio aligned to every frame.
- IMU telemetry
- Accelerometer and gyroscope streams captured with each session.
- Action & object labels
- Verb-noun actions, object identities and hand-object interactions with timing.
- Multi-camera rigs
- Homes, kitchens and industrial settings captured from several viewpoints.
- Custom collection
- Tasks, environments and sensors specified around your robot.
Frame-level label. Values are illustrative.
From head rig to labels.
One clip, through the five stages every recording goes through before it ships.
Capture & consent
Participants wear a head-mounted rig while doing real tasks, with consent covering the recording and the places they record in.
- Who
- Participants
- Gate
- No consent, no footage kept
Privacy
Faces of bystanders, screens and documents are detected and blurred in every frame they appear, then checked by a reviewer.
- Who
- Model + reviewer
- Gate
- Nothing identifying is readable
Sync
Video, audio and motion telemetry are aligned to one clock, so a frame, a sound and a head movement line up exactly.
- Who
- Data engineering
- Gate
- Streams aligned before labelling
Annotate
Actions, objects and hand pose are labelled frame by frame. Failures are labelled on purpose, because slips teach as much as clean grasps.
- Who
- Annotators + reviewer
- Gate
- Reviewer checks every clip
Package
Clips ship with their audio, motion data and annotations in the formats physical-AI pipelines read.
- Who
- Delivery
- Gate
- Metadata per clip
Built for frontier model tasks.
Scope, review criteria and delivery format are agreed per engagement, from the first sample to the final eval.
- Robot manipulation
- Human demonstrations for grasping, placing and tool use.
- Activity recognition
- Fine-grained actions in real workflows.
- Embodied foundation models
- Diverse first-person data for world models.
- Workflow intelligence
- How work is actually done on factory floors.
- Navigation
- Scenes, obstacles and motion patterns for autonomy.
What buyers ask us first.
Something else? Ask a specialist →
01What does a session include?
Head-mounted MP4 video with synchronized audio and IMU telemetry, plus action and object labels scoped to the project.
02Is the footage de-identified?
Faces, screens and bystanders are blurred and rights are documented per session. Additional handling requirements are agreed during scoping.
03Can you collect footage for our tasks?
Yes. We scope tasks, environments, participants and sensor rigs around your robot and agree collection and labelling requirements.
04How does this compare with public egocentric datasets?
Most large public egocentric datasets are academic, released under licence agreements written for research, so commercial use needs checking. Ours is licensed for commercial training, with IMU telemetry synced to each session, action and object labels matched to your robot's action space, and custom collection for tasks the catalogue doesn't cover.
05Do you build benchmarks for physical AI?
Yes. Held-out action and manipulation benchmarks, drawn from licensed sessions and labelled the same way, measure how a model does on real-world footage rather than in simulation.
Same standard in every domain.
Company operations
Workflows, approval trails and legal documents for enterprise agents.
- 01 Dataset
- 52+ companies' ordered approval trails
- 02 Environment
- Agent tasks over approvals, operator-graded
Conversational
Multi-turn conversations, ranked and reviewed for assistant training.
- 01 Dataset
- Multi-turn dialogue, intent-labelled
- 02 Environment
- Multi-turn evals, rubric-graded
Private code
Repositories, commits, pull requests and review history.
- 01 Dataset
- 3B+ lines with commit, PR and review history
- 02 Environment
- Agentic engineering tasks, engineer-graded
Healthcare
Clinical notes, records and imaging, de-identified and expert-reviewed.
- 01 Dataset
- 10M+ de-identified, patient-linked records
- 02 Environment
- Clinical case tasks, clinician-graded
CAD & 3D
Drawings, assemblies and meshes with validated geometry.
- 01 Dataset
- 5M+ typed, native-format CAD files
- 02 Environment
- Part classification and rebuild checks
Audio & speech
Synchronized audio with transcripts and event labels.
- 01 Dataset
- 5.7M+ transcribed, diarized assets
- 02 Environment
- Transcription evals per language and accent
See real egocentric video data first.
A representative sample under NDA, in the format your pipeline reads. Run it through your own evals first.
- Sample records, as delivered
- Datasheet and schema
- Review criteria
- MP4 · AAC · IMU CSV
{"session":"cam018","frame":184,"t":"00:12.8",
"view":"first_person","fps":30,
"action":{"verb":"place","noun":"cup",
"target":"tray","hands":"both"},
"imu":"synced","audio":"aac",
"faces":"blurred","review":"passed"}