Tag
This paper proposes AGWM, an affordance-grounded world model that uses a dynamic prerequisite graph to track action executability in environments with compositional prerequisites. Experiments show it reduces prediction error and improves generalization compared to standard world models.