The Math Behind World Model Paradigms for Robot Learning
A mathematical analysis of three common World Model paradigms in robot learning from the perspectives of probabilistic modeling and structured optimization: IDM-style, Single-backbone, and MoT-style.