Agents
Papers
H-JEPA: End-to-End Learning of Hierarchical World Models for Visual Planning
Zhang et al.
A three-level hierarchy of predictive world models lifts Visual AntMaze success from 18% to 73% with less compute.
In the Daily Brief of 11 October 2026
H-JEPA stacks joint-embedding predictive world models into a hierarchy and plans from the top level down, with higher levels setting goals for lower ones.
A three-level hierarchy raises success on the Visual AntMaze benchmark from 18 to 73 percent while using less compute. It was the most upvoted paper of the week on alphaXiv.