←Home

    Robots That Imagine Before They Act

    How world action models let robots forecast what will happen next as video, points or goal images, and turn that imagined future into actions, from robot-tailored video pretraining and learning from unlabeled footage to long-horizon planning and physically faithful forecasts.

    In order

    1 stories
    機器手拿湯匙, 藍光
    機器手拿湯匙, 藍光

    Photo: Tara Winstead

    Embodied · World action models

    How world action models differ from VLAs, video generators and world models

    A 2026 survey from the National University of Singapore defines robot models that forecast what happens next and act on it, and sorts them along four axes.

    arXiv preprint · 18 June 2026