Humanoid robots are entering mass production, but the AI that drives them is still missing something critical: real-world human data. How people move through a kitchen, hand off a tool, adjust their grip on uneven ground. The kind of physical knowledge that humans absorb through a lifetime of living in bodies, and that no simulation or lab recording can replicate at the scale these models need.

TRACE builds the infrastructure for collecting that data. Our mining partners wear lightweight sensors during their normal daily activities. The data is verified, processed, and added to the WELL (World Experience Learning Library) training dataset. The dataset is licensed to robotics companies and AI labs building the next generation of embodied intelligence. Mining partners own an ongoing share of licensing revenue in proportion to the data they contribute.

We believe the character of robotic systems is shaped by what goes into their training. If that data comes from the full range of human physical culture, the systems that emerge have a real chance of understanding what it means to work alongside people. If it comes from narrow, controlled demonstrations, they will be capable tools and nothing more. We are building toward the first outcome.

The founder writes about these ideas in depth at Bogon Flux on Substack.