Hierarchical graph world model improves zero-shot object navigation success
HGWM: Hierarchical Graph-Guided World Model for Zero-Shot Object Navigation via Graph Matching
RoboticsMultiagent Systems
Summary
Finding objects in new places without prior training is hard for robots. The authors created a method called HGWM that helps robots decide where to look by combining a rough guess of likely locations with detailed memory of what they’ve seen before. Their system scores possible directions and matches object and room information to improve search. Tests in simulated environments show it finds objects more successfully and sometimes more efficiently than previous methods.
What this means in practice
- •For robotics engineers: Design robots that locate objects indoors without prior environment-specific training by combining coarse room priors with detailed observation memory to guide navigation.
- •For simulation developers: Create more realistic benchmarks for embodied AI by integrating hierarchical world models that score navigation choices using language-based priors and spatial graphs.
Tested on simulated data.
Authors
I-tak Ieong, Hao Tang
Abstract
Zero-shot object navigation requires an embodied agent to choose where to search while retaining evidence from previous observations. HGWM couples an LLM-generated room-anchor prior with persistent spatial-semantic memory. A coarse stage assigns visual-semantic scores to candidate headings on a 0-10 scale; a fine stage measures symbolic compatibility between observed objects, room context, and the goal prior. Phase-dependent fusion combines these scores with an LLM direction recommendation before local action selection. On HM3D v0.1, the reported HGWM configuration achieves 62.6% success rate (SR) and 31.5% success weighted by path length (SPL), compared with 58.1% and 31.2% for WMNav. On MP3D, the corresponding values are 46.8% and 17.9%, compared with 45.4% and 17.2%. The results indicate higher success point estimates and modest path-efficiency gains. On HM3D v0.2, the full model obtains 74.4% SR and 31.9% SPL. Additional component ablations reduce SR by 0.9-2.3 percentage points and SPL by 0.6-1.5 points. These are author-reported aggregate measurements. The accompanying implementation repairs pass offline tests; episode-level rerun records are not available. Evaluation is confined to simulation.