Purpose-built egocentric hand video datasets for robotics, embodied AI, and world models. Real workers. Real tasks. Fully licensed.
Thousands of clips across dozens of manual trades — real world, not lab.
Book a Call →First-person 1080p footage. Real workers. No staging, no scripts, no labs.
Every clip goes through a rigorous 4-step pipeline before it touches your training run.
Trained collectors record first-person footage on smartphones mounted to neck or head bands. 1080p minimum. Hands visible. Real tasks only.
Every clip is quality-checked for lighting, hand visibility, task relevance, and privacy compliance. Non-conforming footage is rejected automatically.
Structured metadata: task type, hand actions, object interactions, temporal segments. Your schema or ours — we build the pipeline to your spec.
Training-ready data in your format. JSON, COCO, HuggingFace-ready. Cloud delivery or bulk download. Fully licensed and documented.
Every clip ships with structured metadata. Licensed, real-world footage — not synthetic, not scraped, not staged.
Task type, environment, duration, hand actions, object types.
Action segments with start/end timestamps, grasp phases, idle detection.
No faces. No PII. No children. Every frame reviewed by a human.
JSON, COCO, CSV, HuggingFace. API, cloud bucket, or bulk download.
Every collector signs a data-use agreement. Zero grey area.
Need bespoke labels? Depth data? We build to your spec.
We do one thing — egocentric hand video. We do it better than anyone.
No actors. No scripts. No controlled environments. Our footage comes from real workers doing real tasks in actual homes, workshops, and kitchens across multiple countries.
500+ trained data collectors and growing. Suburban kitchens. Market stalls. Tailor shops. City workshops. Not a lab — the real world at scale.
Need 100 hours of kitchen data? 5,000 hours across 8 domains? Our network scales with your training needs. From brief to delivery in days, not months.
No faces. No PII. No minors. Collection protocols built for compliance from day one. Every clip reviewed frame-by-frame before delivery.
From brief to first delivery in days, not months. We scope the dataset, design the capture protocol, and deliver training-ready data on your timeline.
Or email us directly at data@handsframe.com