World Action Models give robots the ability to simulate consequences before they move

World Action Models address a fundamental weakness in robotics AI, where current models learn movement-image correlations but fail to grasp environmental changes. A new survey organizes around a hundred papers into two architectural lines, revealing a key advantage: these models can learn from everyday videos without robot action labels. Such data was nearly useless for traditional robotics AI, indicating that these new models could significantly enhance robots' understanding of their surroundings.