Event
HGWM: Graph-Guided World Model Achieves Zero-Shot Object Navigation
1 reports · 1 sources
I-tak Ieong and Hao Tang released HGWM, a method for zero-shot object navigation: it combines room-anchor priors generated by a large language model with persistent spatial-semantic memory to select search directions. The authors report a success rate of 62.6% and SPL of 31.5% on HM3D v0.1, compared with WMNav's 58.1% and 31.2%; on MP3D, 46.8% and 17.9%, compared with 45.4% and 17.2%. These are author-reported aggregate figures, and evaluation is limited to simulated environments; the accompanying implementation passes offline tests, but there is no per-episode rerun record.
Report timeline
- HGWM: Zero-Shot Object Navigation Using Hierarchical Graph-Guided World Models and Graph Matching
The authors propose HGWM, which combines an LLM-generated room-anchor prior with persistent spatial-semantic memory to select search directions in zero-shot object navigation. On HM3D v0.1, it reports 62.6% success rate and 31.5% SPL, versus 58.1% and 31.2% for WMNav; on MP3D, 46.8% and 17.9%, versus 45.4% and 17.2%.
arXiv cs.RO ·