Questions, answered.

We capture how master practitioners do real physical work, with consent: where their attention goes, the judgment behind each move, what happens at the point of contact, and whether the work succeeded. The full breakdown of what each session records is shared with labs under NDA. The point is to capture the judgment behind the action, not just the action, including the moments where an expert notices a problem and recovers.

Every dataset is documented to a consistent schema that covers its provenance, the consent behind it, and how a session is packaged for training, with exports to the formats robotics teams already use. We keep the full specification private for now and share it with labs under NDA, with plans to publish it openly in 2026. Every session is processed on Loom, our internal capture-to-corpus pipeline, and we plan to open parts of it to the community in time.

We are early. Three verticals are signed and scheduled for capture across nine partner sites, starting with Hospitality F&B, and first captures begin August 2026. An illustrative EDS schema and episode manifest are available under NDA, and live evaluation samples will follow the first capture sessions. If you have a specific vertical or task in mind, talk to us and we will share what is feasible and on what timeline.

Consent is explicit, scope-limited, and opt-in by default, and practitioners can withdraw and request deletion at any time. Practitioners are paid above-market session rates with royalty participation on commercial licensing. There is no piece rate, and practitioners retain authorship of their work.

Capture is aligned with India's DPDP Act 2023. Data is India-resident and cold-stored with an audit-grade chain of custody recorded at the frame level. Buyer briefings can include further detail under NDA.

Practitioner identities are hashed in records and protected under named-consent and licensing agreements rather than published. Eye gaze is delivered as derived feature vectors only, not raw imagery of a person's eyes, so attention can be learned without exposing identity.

Every dataset is documented to a consistent schema and exports natively to LeRobot, Open X-Embodiment, and RLDS, so it drops into the pipelines the robotics community already uses.

The schema is designed for retargeting to standard arm and humanoid classes. Per-embodiment delivery is scoped per engagement, so we can match the export to the platform you train on.

Teleoperation captures a robot being driven, and simulation captures a model of the world. Anthrobase captures a human expert doing the real task, with their gaze, force, and judgment aligned to the outcome. That gives models the action priors and recovery signals that neither teleoperation nor simulation contains.

Sessions are reviewed with human-in-the-loop quality assurance. Named technical reviewers, currently from IIT Delhi and IIT BHU, audit synchronization and stream quality and tag decision points and tacit knowledge against the schema before a record is signed.

Yes. Tell us the vertical, task, embodiment, and sample budget, and we respond within one working day with a feasibility note and a timeline. Custom capture is the core of what we do.

Yes. We partner with hotels, hospitals, MRO facilities, craft collectives, and training institutes. We supply the hardware, the operators, and the legal and consent frameworks. Reach out through the contact page to start a conversation.