Native Action-Prior Learning from Videos for World Action Models

Zhaochong An, Fei Zhang, Menglin Jia +10

World action models integrate future visual dynamics with robot action prediction, but their scalability remains limited by the need for action-annotated robot trajectories. Observ…