Reproduce “H-JEPA: End-to-End Learning of Hierarchical World Models for Visual Planning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wancong Zhang, Basile Terver, Michael Rabbat, Yann LeCun, Randall Balestriero
H-JEPA plans visual maze routes with three levels that predict at different temporal scales in separate latent spaces. A coarse planner proposes subgoals, and finer levels refine them into actions. In simulated AntMaze, it achieves 73.3% success (±3.5 SE), versus 18.0% (±3.5 SE) for flat LeWM across 50 held-out tasks, under a selected planning-compute budget.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.