FAQ / TRUELABEL
Frequently asked questions.
Buyer-facing answers for procurement, ML, and robotics leads evaluating truelabel as a sourcing partner for physical AI training data. If a question is not answered here, the team will answer it directly — contact details are at the bottom of the page.
Questions
What is truelabel?
truelabel is a data marketplace for physical AI. Robotics, embodied-AI, and VLA teams send one spec — robot, task, environment, modality, license — and truelabel routes it to candidate capture suppliers reviewed against the buyer spec who collect egocentric video, teleoperation traces, manipulation demos, and field datasets. Deliveries arrive rights-cleared with provenance, consent artifacts, and per-trajectory metadata attached.
Is truelabel the same as truelabel.io, truelabel.com, or other companies with the name?
No. truelabel.ai is unrelated to consumer-product label verification services, supply-chain traceability companies, or any other business using a similar name. truelabel.ai operates exclusively as a physical AI data marketplace for robotics, embodied AI, and foundation-model teams. The official domain is truelabel.ai.
See:About truelabel
How does the marketplace actually work, from spec to delivery?
A buyer writes a spec at /specs covering platform, task, environment, modality, volume, and license. truelabel fans the spec to qualified capture partners. Sample packets — small, labeled, rights-cleared, QA evidence attached — come back first. The buyer approves the suppliers that match, then scales. Final deliveries ship in the requested format with provenance and consent intact.
What modalities does truelabel cover?
Egocentric video (head-mounted, wrist, scene), teleoperation traces (leader-follower, VR/exoskeleton), manipulation demos (single-arm and bimanual), mobile manipulation, locomotion, and field datasets captured on production sites. Multimodal capture — RGB plus depth, force-torque, joint states, gripper telemetry, language annotations — is supported through partners who instrument for it.
See:Modality catalogBriefings on teleoperationEgocentric briefings
What formats do you deliver?
RLDS, LeRobot v2, MCAP, ROS 1 / ROS 2 bags, HDF5, Parquet, and per-buyer custom layouts. Format is part of the spec, not an afterthought. Schemas, episode segmentation, and per-trajectory metadata are validated against the buyer's training pipeline before scale.
How do you verify capture partners?
Every partner is vetted against a published quality bar before they receive specs: rig inventory, calibration evidence, capture history, consent infrastructure, jurisdiction, and prior delivery references. Sample packets are required before any scaled engagement. Partners who fail QA on samples do not receive volume work. The bar is enforced per-spec, not just at onboarding.
Who owns the data rights once it's delivered?
Rights are negotiated per engagement and stated in the spec. Model training, derivative works, downstream deployment, exclusivity, retention, redistribution, and resale are separate terms to review rather than assumed defaults. Rights-cleared delivery includes contributor consent artifacts, location releases where applicable, and per-trajectory provenance/metadata for buyer review.
What about consent and provenance for captured data?
Captured-data review keeps contributor consent artifacts, location releases where applicable, and per-trajectory provenance/metadata attached to the delivery. For real-home demonstrations, bystanders and private spaces remain separate review questions; an artifact's presence does not by itself establish legal sufficiency for every use or jurisdiction.
How is pricing structured?
Pricing is bespoke. Physical AI data cost varies by modality, rig, geography, operator skill, hours per trajectory, post-processing, and license. truelabel quotes per spec after partner matching, not from a public rate card. The cost estimator tool gives a directional range; final pricing is settled with the partner before capture begins.
What size buyers does truelabel work with?
Frontier foundation-model labs, embodied-AI startups, robotics OEMs, and academic groups with industrial budgets. The marketplace is built for buyers who need sourcing rigor — provenance, consent, license clarity, format conformance — not just bodies in a room. Engagement size ranges from single-spec pilots to multi-quarter, multi-partner sourcing programs.
How long does a typical sourcing engagement take?
Spec to first sample packet is usually one to three weeks, depending on rig availability and capture geography. Scale starts after the buyer approves samples. A standard production engagement runs four to twelve weeks; long-running collections, multi-site programs, and partner co-development engagements run longer and are scoped at signing.
Do you support custom rigs or capture vendors I already know?
Yes. Buyers can bring named partners into the marketplace — truelabel runs them through the same verification gate and operates the engagement under the standard QA, provenance, and rights framework. Custom rigs (proprietary teleop hardware, instrumented gloves, multi-camera arrays) are supported when the partner can document calibration and capture telemetry to spec.
Have a question we didn't answer?
Send it to hello@truelabel.ai and a human on the team will answer. If you are ready to send a spec, start at /sourcing. If you operate a capture rig or run a data collection studio, apply at /apply.