World action models (WAMs) are policy models that explicitly predict both the next action and the future world state. The abstract of a new arXiv paper notes that WAMs have been shown to outperform action-only models across various robotic manipulation tasks.

The paper, titled "Box-WAM," proposes using box trajectories—sequences of bounding boxes—as a compact state representation for these models. This is presented as a way to make the predicted world states more efficient to represent, though the abstract excerpt does not specify the exact limitations of prior WAMs that motivated this design.

The source is a single arXiv abstract, and the text cuts off mid-sentence after "However," so the full problem statement and experimental results are not available in the provided material. The claims here are limited to what the abstract explicitly states.