World-as-Graph: Relational World Modeling Through Latent Space Graphs
The proposed World-As-Graph (WAG) is a graph-based object-centric world model that introduces relational inductive bias into JEPA-style predictive representation learning and could demonstrate the superior performance of the proposed WAG on both visual reasoning and robotic manipulation tasks.
Ya-Qi Yang, Shu-Wei Huang, Yu-Jin Huang et al.
· 1 citation