Kinetic Blocks opened its marketplace for commercially licensed, provenance-tracked robot training data in beta on 1 September. The first month has taught us more about how this market works than the planning we did before launch.
The first lesson concerns balance. Roughly nine out of ten organisations registering on the platform are sellers. They range from capture teams recording egocentric video in factories, warehouses and construction sites to teleoperation operators and integrators who collect data as a by-product of deployment. Buyers make up the remaining tenth. The capacity to produce robot data is growing faster than the purchasing processes at the companies training robot policies.
The second lesson is that every transaction still runs manually, because people on both sides still make the decisions about data purchases. Sellers and buyers find each other, exchange samples, agree terms and arrange delivery over email and Teams. No buyer will yet let a system commit budget to a supplier it has never met, and no seller will hand over a dataset because an algorithm says the counterparty is good for the money.
What we hear from the two sides tells us why. Buyers ask first about quality. Does the data cover the tasks, environments and hardware their models need? Do the annotations hold up under inspection? Will the consent and licensing documentation survive legal review? Then they ask who we are, and that question is just as reasonable for a platform that did not exist two months ago. Our answer is to score quality at the point of capture, document the chain of custody for every dataset and have sellers upload samples to their own accounts on the platform, so that buyers can judge the material themselves before they commit. Sellers ask a different question: whether they will be paid, and when. Data cannot be returned once it has been delivered, and that makes sellers cautious with a new counterparty. In China, where suppliers commonly take payment in advance, establishing the same platform-held payment terms will take longer than in other markets, and we intend to put in that work.
Trust of this kind takes time to build. Trade becomes automatic only when there is a record of completed transactions in which both sides received what was promised. We are the first truly open marketplace dedicated to licensed robot training data, so there is no track record in this category to borrow from, and we are building it one transaction at a time. It helps that the people we have met on both sides have generally been eager to get started and good at what they do.
We also want to be open about what we do not yet know. Our quality scoring tells a buyer how a dataset performs on capture and documentation. Whether that data is good enough to train a working robot policy is still an open question for us, as it is for the rest of the market. We therefore support the work others in the field are doing to benchmark training data against policy performance, and we are planning a benchmark of our own that will connect our scoring to outcomes in robot training.
We will be in San Francisco for Tech Week from 5 to 11 October, and we would like to meet four groups there: buyers training robot policies, sellers with data to list, integrators holding deployment data and investors following physical AI. Get in touch through kineticblocks.com so we can find a time during the week. Data is digital, trust is still human, and that is why we would like to meet you in person.