Each anchor view provides an RGB image, a 3D pose for spatial grounding, and an evolution prompt that localizes how the scene should change. The method combines hybrid-view human action control, evolvable anchor-view customization, and progressive training to generate personalized egocentric environments.
AnchorWorld is useful for embodied AI, game-like simulation, egocentric video generation, and interactive world-model research. The page links to a paper and dataset, while code availability is not clearly shown in the submitted page.
