SSR-ZSON: Zero-Shot Object Navigation Via Spatial-Semantic Relations within a Hierarchical Exploration Framework
Xiangyi Meng, Delun Li, Zihao Mao, Yi Yang, Wenjie Song
AI summary
Problem
Existing zero-shot object navigation methods struggle with inefficient exploration due to weak semantic guidance and frequent local entrapment from poor spatial memory in unknown environments.
Approach
The framework discretizes environments into grid sub-regions and uses an LLM to dynamically score semantic associations, guiding the robot to prioritize unexplored areas with high target-relevance while balancing geometric coverage.
Key results
- 18.5% and 11.2% Success Rate gains on Matterport3D and Habitat-Matterport3D
- 0.181 and 0.140 SPL improvements over state-of-the-art methods
- Real-time operation on hybrid simulations and physical platforms
- Novel viewpoint-region joint navigation framework with dynamic semantic weighting
Why it matters
Provides a practical, training-free navigation solution for robots operating in complex, unstructured environments without prior maps.
Abstract
Zero-shot object navigation in unknown environ- ments presents significant challenges, mainly due to two key limitations: insufficient semantic guidance leads to inefficient exploration, while limited spatial memory resulting from en- vironmental structure causes entrapment in local regions. To address these issues, we propose SSR-ZSON, a spatial-semantic relative zero-shot object navigation method based on the TARE [1] hierarchical exploration framework, integrating a viewpoint generation strategy balancing spatial coverage and semantic density with an LLM-based global guidance mechanism. The performance improvement of the proposed method is due to two key innovations. First, the viewpoint generation strategy prioritizes areas of high semantic density within traversable sub-regions to maximize spatial coverage and minimize in- valid exploration. Second, coupled with an LLM-based global guidance mechanism, it assesses semantic associations to direct navigation toward high-value spaces, preventing local entrap- ment and ensuring efficient exploration. Deployed on hybrid Habitat-Gazebo simulations and physical platforms, SSR-ZSON achieves real-time operation and superior performance. On Matterport3D [2] and Habitat-Matterport3D [3] datasets, it improves the Success Rate(SR) by 18.5% and 11.2%, and the Success weighted by Path Length(SPL) by 0.181 and 0.140, respectively, over state-of-the-art methods.