Reproduce “HGWM: Hierarchical Graph-Guided World Model for Zero-Shot Object Navigation via Graph Matching” 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

I-Tak Ieong, Hao Tang

HGWM searches for category targets in unfamiliar homes by proposing rooms, anchor objects, and relations, while keeping those hypotheses separate from observed evidence. It compares the two to rank directions, using spatial-semantic memory and model-based scores. On HM3D v0.1, it reports 62.6% success and 31.5% SPL, versus WMNav’s 58.1% and 31.2%; the gain is clearer in completion than path efficiency.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.