NVIDIA publishes analysis on the evolution from Vision-Language-Action (VLA) models to World Action Models (WAM), systems that integrate pretrained models to generate complex robotic behaviors based on visual observations and natural language instructions.