The most human training data
Egocentric, full-recipe cooking video from real home kitchens — anonymised, annotated, and ready to train household and manipulation robots.
Real kitchens, captured first-person.
Every clip is a complete recipe cooked start to finish in a real home — the messy, varied, long-horizon manipulation that lab setups can't reproduce.
First-person RGB video
Head-mounted, egocentric footage at up to 4K — the robot's-eye view of every action.
Depth & LiDAR
Premium captures add synchronized depth maps and LiDAR from iPhone Pro devices.
Synced audio
Time-aligned audio for sizzles, timers and spoken context around each step.
Rich annotations
Action segments, step boundaries, and object & ingredient tags on every clip.
A transparent pipeline, end to end.
From the home cook to your training bucket — every step is consented, documented, and reproducible.
Sourced
Real home cooks, recruited and fairly paid — never actors or staged sets.
Captured
Full recipes filmed first-person in their own kitchens via the Fogón app.
Anonymised
Faces and personal details are blurred automatically in the processing pipeline.
Annotated
We label steps, actions and objects, then quality-check every clip.
Delivered
Curated train/val splits in your schema, to your cloud bucket.
Data you can defend in a review.
Every clip is sourced with informed consent from fairly-paid cooks, anonymised automatically, and handled under EU data-protection law.
Informed consent & fair pay
Every cook opts in explicitly and is paid for their work. No scraping, no surprises.
GDPR compliant
Collected and processed under EU data-protection law, with a lawful basis per clip.
Automatic anonymisation
Faces and PII are blurred by default — your model never sees an identifiable person.
Deletion on request
Contributors can withdraw, and we propagate deletions to delivered datasets.
Never sold to advertisers
The footage exists only to train robots. It is never used for ads or surveillance.
Auditable provenance
Each clip carries consent and processing metadata you can show in a model review.
Dataset at a glance.
- Modalities
- RGB video, depth/LiDAR (premium), audio, annotations
- Resolution
- Up to 4K, 30–60 fps
- Format
- MP4 / H.264 + depth maps
- Annotations
- Action segments, step boundaries, object & ingredient tags, transcripts
- Environment
- Real home kitchens
- Diversity
- 60+ cuisines across Spain & the EU
- Consent basis
- Per-clip informed consent, GDPR
- Licensing
- Commercial & custom terms
- Samples
- Datasheet & sample on request
- Delivery
- Cloud bucket, curated splits, your schema
Figures are indicative and grow continuously. Ask for a current datasheet and sample.
Made for embodied AI.
Manipulation policies
Long-horizon, contact-rich manipulation grounded in real human demonstrations.
Household robots
Teach robots the everyday kitchen tasks people actually do at home.
Vision-language-action models
Pair video with step text and action labels to train VLA and world models.
Action & affordance recognition
Densely segmented actions and objects for recognition and affordance learning.
Imitation learning
Thousands of complete demonstrations per skill, from the actor's viewpoint.
Benchmarking
A diverse, real-world test set for evaluating embodied and cooking-specific models.
Let's talk data.
Tell us about your lab and what you're training. We'll reply with a datasheet, sample, and licensing options.
or email us directly at datasets@fogon.ai