CSG-Nav: Confidence-Aware Semantic-Geometric Fusion for Robust Exploration and Accurate Zero-Shot Object-Goal Navigation
CSG-Nav: Confidence-Aware Semantic-Geometric Fusion for Robust Exploration and Accurate Zero-Shot Object-Goal Navigation
This research presents CSG-Nav, a zero-shot navigation framework designed to enable autonomous agents to explore unseen environments and locate target objects without task-specific training. The system addresses challenges in perceptual reliability under observation uncertainty by integrating three complementary fusion mechanisms: semantic-source fusion to suppress hallucination-induced directional bias, spatial-confidence fusion to weight multi-view evidence using Gaussian decay, and semantic-geometric decision fusion for physically grounded stopping decisions. Experiments on HM3Dv1, HM3Dv2, and MP3D datasets demonstrate competitive zero-shot performance with success rates of 58.9%, 71.6%, and 44.3% respectively. The framework is validated through fusion-level ablations and real-world deployments in both indoor and outdoor environments, with code available for public use.
The paper introduces CSG-Nav, a zero-shot navigation framework that fuses semantic and geometric data to mitigate hallucination and improve stopping accuracy in unseen environments.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Control
- Reported autonomousOpen source S2
Reported numbers
Reported success rate
58.9%
Reported zero-shot result; the excerpt does not specify an evaluation protocol.
Reported trials
View original evidence
ion combines semantic target detection with depth-based 3D localization to support physically grounded stopping decisions. Experiments on HM3Dv1, HM3Dv2, and MP3D demonstrate competitive zero-shot performance, achieving success rates of 58.9%, 71.6%, and 44.3%. Fusion-level ablations and real-world deployments in indoor and outdoor environments further verify the effectiveness and practicality of CSG-Nav for robust n
Open source S5
- code: Available Artifact link Open source S6
What remains unknown
Not established in the collected evidence: Environment, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
arXiv:2611.11978v1 Announce Type: new Abstract: Zero-shot object-goal navigation requires autonomous agents to explore unseen environments and locate target objects without task-specific training, posing challenges in perceptual reliability under observation uncertainty and physically grounded goal termination. Although foundation-model-based approaches improve generalization, they remain vulnerable to unreliable lon
Open source S2
ion combines semantic target detection with depth-based 3D localization to support physically grounded stopping decisions. Experiments on HM3Dv1, HM3Dv2, and MP3D demonstrate competitive zero-shot performance, achieving success rates of 58.9%, 71.6%, and 44.3%. Fusion-level ablations and real-world deployments in indoor and outdoor environments further verify the effectiveness and practicality of CSG-Nav for robust n
Open source S5
avigation under uncertain perception. Code is available at https://xr-ui.github.io/CSGNav.github.io/
Open source S6
Source：arXiv Robotics — research abstracts · arxiv.org