Atomic-WAM: Execution-Grounded Atomic-Motion Reasoning for World Action Models

This paper presents Atomic-WAM, an understand-then-execute framework that connects task reasoning to action generation through atomic-motion descriptions. The approach uses a vision-language reasoner to describe next local motion from task instructions and visual context, guiding a World Action Model (WAM) to predict future video and action chunks. An automatic annotation pipeline curates temporally aligned, contact-enriched supervision from multi-view DROID demonstrations. The reasoner is post-trained with Execution-Grounded GRPO (EG-GRPO) to adapt through rewards for motion appropriateness and execution success. Atomic-WAM achieves 39.6% task success under default instructions and 31.6% under vague instructions in a real-to-sim evaluation on RoboLab.