Egocentric video datasets for robotics manipulation training.

Most modern container/codec combinations: MP4 (H.264, H.265), MOV, MKV, WebM. Resolution from 720p to 4K. We strongly prefer egocentric footage (head-mounted camera, smart glasses, or chest-mounted GoPro), but we can work with first-person teleop or wrist-camera footage if that is what your project requires.

Yes. Natural language narration is a collection modality in its own right, alongside egocentric video. Contributors describe what they are doing as they do it. Narration is captured as a separate audio track and can be requested on any collection campaign. Voice is licensed for model training only. Voice cloning and speech synthesis are excluded.

Timeline depends on volume and engagement model. For annotation on a batch of existing video, a first sample comes back in days. For collection plus annotation, we agree a delivery date on your actual scope at kickoff and ship the QA report with the data. We would rather commit to a date on your project than publish an average that means nothing.

Three layers. Faces are blurred before any external processing; this is a minimisation measure and it does not make the footage anonymous, which we state plainly rather than overclaim. Storage and the main processing run in the EU, and the residual transfers arising from suppliers established outside the EEA are covered by the European Commission's standard contractual clauses, each one listed with its supplier on our sub-processors page. Every processing activity has a documented legal basis, and a contributor can ask for erasure at any time, which we action within thirty days.

Yes. Tell us the task, the environment, and any constraints on objects, lighting, or contributor profile. Our contributor network is anchored in Europe and open worldwide, and we brief the contributors whose setting matches your scenario, filtering on country of residence where a project requires it. For a task nobody has captured yet, we design the mission brief around it rather than adapting existing footage.

LeRobot v3.0 is the Hugging Face robotics dataset format: parquet shards plus MP4 frames plus JSONL metadata, native to the LeRobot training stack. RLDS is the TFRecord-based format used by Open X-Embodiment and the broader TFDS ecosystem. We ship both from the same source data; pick whichever matches your training pipeline.

Annotations. Our hand pose, depth, segmentation, and action labels are model predictions, not ground truth. They have measurable accuracy but are not human-verified frame by frame. Every dataset ships with uncertainty metadata so your team can filter or weight noisy samples. If your evaluation needs human-verified labels on a subset, tell us at kickoff and we will scope it as a separate deliverable.

The QA report ships before the dataset is final. If a clip does not meet the specification agreed at kickoff, we replace it, and where replacement is not possible we refund the corresponding portion of the fees. We also support evaluation pilots, a small sample before scaling, so you can validate quality on your own benchmark.

Yes. We have a standard mutual NDA template that covers your project specifics, training objectives, and the contents of the dataset. We can sign it before the first technical call. Email contact@egovista.app and we can have it back to you within 24 hours.