World Action Planner Uses VLM Reasoning for Robot Planning

World Action Planner combines Vision-Language Model reasoning with a multi-task pose-image conditioned world model to enable iterative action refinement. The system outperforms end-to-end VLA and WAM models in zero-shot generalization and compositional tasks by reasoning over imagined rollouts.