The company is positioning this approach as a turning point for robotics, comparable to what large generative models have done for text and images.