Robot Training Data Quality Survey Please enable JavaScript in your browser to complete this form.Please enable JavaScript in your browser to complete this form.Kinetic.blocks is building infrastructure for commercial robot training data. We are conducting a short research survey to understand how leading teams currently manage data quality requirements and vendor relationships. This takes 5–7 minutes. Responses inform our product roadmap and a forthcoming industry report on humanoid training data standards, published on humanoid.guide. Your responses can be anonymous or attributed – your choice. SECTION 1 – YOUR ROLE Q1. What best describes your organisation? *Humanoid robot OEMFoundation model / world model teamData collection operationAI cloud / infrastructure providerResearch institutionOther:Other:Q2. Are you currently sourcing robot training data externally – from vendors or marketplaces? *Yes, regularlyYes, on a project basisEvaluating optionsNo, we collect everything in-houseSECTION 2 – QUALITY REQUIREMENTS Q3. Do you have a formal quality specification for egocentric or humanoid robot training data? *Yes, a detailed written specYes, informal / undocumentedNo, we evaluate case by caseNot yet, but we are developing oneQ4. Would you be willing to share your quality specification with a potential data vendor?Yes, openlyYes, under NDAOnly after an extended period of collaborationNo – it is proprietary IPDepends on the vendorQ5. What are the most important quality dimensions for your organisation? Rank top 3: *Temporal integrity (frame continuity, sampling frequency, sensor sync)Signal quality (camera stability, action space cleanliness)Annotation depth (clip-level, subtask, primitive labels)Provenance and licensing clarityDataset diversity and completenessFormat compatibility (LeRobot, RLDS, ROS bag)Price per validated hourOther:Other:SECTION 3 – VENDOR ONBOARDING Q6. What does your current process look like for onboarding a new external data vendor?We evaluate sample data firstWe require a pilot delivery before committingWe start with an NDA and then share specsWe do not have a formal processOther:Other:Q7. What is your typical timeline from first contact with a new vendor to first data delivery?Less than 2 weeks2–6 weeks1–3 monthsMore than 3 monthsWe have not done this yetQ8. What would make you more likely to work with a new, unknown data vendor? Select up to 3:Third-party quality certification (e.g. KBQS)Sample data that meets your specReference from a known organisationCompetitive pricingProven ability to deliver at volume (10 000+ hours)Compatible format out of the boxCommercial license with clear IP termsSECTION 4 – INDUSTRY STANDARDS AND PRICING Q9. Is there a need for an industry-wide quality standard for robot training data — or must quality requirements always be customised per world model or robot platform? *A shared baseline standard is valuable, with customisation on topFully customised per platform – a shared standard is not practicalNot sure yet – the field is moving too fastA standard would be valuable but needs to come from a neutral bodyQ10. If a trusted quality standard existed, would your organisation pay a premium for certified data?Yes, meaningfully more (>20%)Yes, somewhat more (5–20%)No price premium, but it would simplify vendor selectionNo – we would always verify quality ourselvesQ11. What is your current price range for externally sourced egocentric video data?Under $30/hour$30–50/hour$50–100/hour$100–200/hourOver $200/hourWe do not currently purchase this type of dataQ12. What is your current price range for high-quality robot teleoperation data?Under $100/hour$100–200/hour$200–300/hourOver $300/hourWe do not currently purchase this type of dataSECTION 5 – OPEN QUESTIONS Q13. What is the single biggest gap in the current robot training data market that nobody is solving well?Q14. Would you be open to a follow-up conversation about data quality requirements and vendor onboarding? *YesNo What Select or Name *Email *Job title *OrganizationSubmit
