Language models had the internet. Robotics needs an internet equivalent. We believe demonstrations collected through teleoperated robots alone cannot provide the scale and diversity needed for robots and humanoids to learn the physical world. People already navigate that world and manipulate objects every day. Egocentric data, first-person video of human demonstrations, offers a way to capture that experience at scale. The goal is to enable robots to learn from people across a vast diversity of tasks, environments, and countries. CR AI Labs is building toward that future: human experience as a foundational learning resource for robotics.
