The robotics industry is racing to teach machines to operate in the physical world, but a lack of training data is holding them back. XDOF, a new startup, is addressing this gap by building a data ecosystem for companies pursuing robotics models. The company has raised $70 million and partnered with UC Berkeley to release a large collection of high-quality robot training data. This move is significant as it could help frontier labs and robotics companies avoid falling behind in the language model race and enable the development of more capable robots. By providing a self-reinforcing feedback loop for robot trainers, XDOF is creating a new kind of infrastructure business that could have a major impact on the industry.